Frequencies and box plots of resemblance rating responses (scores out of 10) for the facial approximation constructed by the first author (CNS). In the box plots, the rectangle indicates the inter quartile range (25–75th quantiles), with the middle line representing the median. The diamond, at its highest point, indicates the arithmetic mean; the left and right tips of the diamond represent 2 standard errors or the 95% confidence interval. The ‘‘whiskers’’ or lines coming off the rectangle represent: the upper quartile + (1.5 Â interquartile range); and the lower quartile À (1.5 Â interquartile range). The bracket at the edge of the interquartile box represents the ‘‘shortest half’’ or the most dense 50% of the observations. Kurtosis values have been rescaled relative to 0, not 3. 

Contexts in source publication

Context 1

... Nine other distractor faces of the same sex, approximate age and approximate pose as the target individual were selected from Australian newspapers and included in the face array. All images were resampled 1 down (pixels removed), scaled and cropped in Adobe 1 Photoshop 6.0 to give images as closely comparable as possible to each other (Fig. 3). No effort was made to select individuals of similar physical appearance (in this case facial appearance) to the target as recommended in usual eyewitness tests. Thus, this face array consisted of a fairly random sample and is expected to provide a scenario favourable for correct recognitions of the facial approximation (the face array may be biased and hence some distractor faces may not be seen to be plausible alternatives to the target individual and/or the facial approximation). To determine if any type I bias was present in the face array, sequential and simultaneous presentation trials were conducted without the facial approximation. Fifteen adult assessors (11 females, 4 males: mean age 23 years, standard deviation 12 years) who did not recognize, either personally or as having been seen in the media, any of the faces in the face array were asked to determine, without the facial approximation, who the murder victim (target individual) was in the face array. Assessors had no information apart from the photographs themselves to base this decision on. Assessors first participated in the sequential face array presentation and then the simultaneous line-up. In the sequential face array assessors where shown one face image at a time in a random order. Assessors were forced to decide for each face ‘‘yes’’ or ‘‘no’’ if the face was that of the murder victim. Assessors were aware that if they made an identification of ‘‘yes’’ the trial was completed (they would not see the other faces in the sequence during this test) and if they answered ‘‘no’’ that they would not be able to change their mind to this face at a later point in the trial. The investigator (CNS) held more cards than those included in the face array test so that assessors could not anticipate the end of the sequence. In the sequential trial cards were held about 1 m in front of the assessor and at arm’s length from the investigator at 90% to his line of sight so investigator cues, if there were any, were not obvious to assessors (who were hopefully concentrating on the cards and hence not attending to any cues, particularly facial ones, if expressed by the investigator). After assessors had made an identification, or if they proceeded through all the faces in the sequence without making an identification, all the faces in the array were presented simultaneously, but in a random order, to the assessor for him/her to change their identification decision from the sequential trial if they so wished. Twenty new assessors (14 females, 6 males: mean age 20 years, standard deviation 4 years) who did not recognize, either personally or as having been seen in the media, any of the faces in the face array, were recruited for the main project: to attempt to correctly identify who the target individual was from the facial approximation. Identical procedures were followed as indicated above for face pool testing without the facial approximation, except of course that this time the assessors had access to a facial approximation. Although assessors had the option of not choosing any face in the simultaneous line-up, a chance rate of 10% was used as it was found that almost all assessors selected a face (see Section 3), and hence appeared biased in this respect (according to chance one would expect a 50% response of ‘‘not there’’, instead it seemed all assessors were choosing a face and hence each face in the array had a 10% chance of being selected). Responses were recorded categorically as correct (target face identification) or incorrect (distractor identification or ‘‘no identification’’ response) for statistical analysis. Observed data were compared to expected frequen- 1 cies by Fisher’s Exact tests in the JMP 4.0 statistical package. As we were only interested in responses larger than chance (and because chance rates were so small that lesser differences would be difficult to detect) confidence intervals for one tailed tests were used. Assessor’s resemblance ratings of both facial approximations were in accordance with previous indications from other individuals, including other forensic facial approximation experts, that resemblance was high. It is worth noting that Betty Pat Gatliff, a high profile forensic artist and facial approximation practitioner, indicated to the first author in 2000 that the face appeared similar enough that she would have expected a positive result had it been advertised. Both facial approximations received high, but similar, resemblance rating results (around 7 out of 10), although the facial approximation without hair tended to be rated higher than the facial approximation with hair (Fig. 4). Data distributions for the no hair facial approximation displayed slightly more left skew than those for the facial approximation with hair (Fig. 4). Hence in recognition tests reported here we used the facial approximation without hair as this is expected to favour positive recognition responses according to traditional facial approximation theory (greater resemblance, greater accuracy and possibly more frequent correct recognitions). Sequential and simultaneous face array tests, done without assessors seeing a facial approximation, showed that the target individual (face number 4) was identified at rates well above other individuals (Fig. 5) and that responses were highly similar between the sequential and simultaneous presentation methods. For sequential line-ups, the identification of the target individual (face number 4) in comparison to other faces was statistically significant ( p < 0.08) in six out of nine cases (i.e., for face numbers 1, 2, 5, 6, 7, and 8), although the other three cases (face numbers 3, 9, and 10) followed similar trends. In no case was a ‘‘not there’’ response made for either the sequential or the simultaneous face array. During post-experiment feed back assessors expressed their thoughts why they thought they could tell who the murder victim was without the facial approximation. Frequently these explanations included: ‘‘the target photograph . . . would not be flattering’’, ‘‘ . . . would not be clear’’, and/or ‘‘ . . . would present the person smiling’’. When the facial approximation without hair was presented to assessors who then attempted to identify the target individual from the sequential face array, identification of the target individual (face number 4) did not increase (Fig. 6). Face number 4 was not identified above chance rates at statistically significant levels. Face number 4 was also identified some- what less, but not at rates statistically different from that observed when the facial approximation was not presented to assessors (in both sequential and simultaneous procedures). During sequential trials the majority of responses were the default response of ‘‘no identification’’, as most individuals (60%) completed the sequential presentation without identifying any face. The increased number of ‘‘not there’’ responses for the sequential face array, facial approximation present scenario, was highly statistically significant ( p < 0.001) in comparison to: (i) the sequential face array, facial approximation not present scenario; (ii) the simultaneous face array, facial approximation not present scenario; and (iii) the simultaneous face array, facial approximation present scenario. Thus, when facial approximations were presented to assessors in the facial approximation present scenario, the number of false identifications decreased in comparison to all other testing conditions. The similarity ratings of the facial approximation to the target individual clearly indicate high resemblance, as the modes observed for the facial approximations were high (7 out of 10 for the facial approximation with hair; and 6/10 and 8/10 for facial approximations without hair; Fig. 4). Such resemblance ratings seem comparable with reports of resemblance ratings for other ‘‘successful’’ facial approximations [14–17]. However, despite attempts to encourage ‘‘favourable’’ recognition results here (by using the facial approximation that achieved the highest resemblance, and not selecting a face pool that included distractors who looked like the target individual) recognition frequencies of the target individual were poor, and in fact tended to be less than those when assessors made an identification without the use of the facial approximation . Furthermore, when assessors used the facial approximation in sequential trials the ‘‘no identification responses’’ increased dramatically in contrast to responses when facial approximations were not used. These results offer strong support for claims that resemblance ratings do not indicate the recognizability, and hence accuracy, of facial approximations [2], and that facial approximations are not recognized frequently or reliably above rates expected by chance [1]. Results were also consistent with eyewitness research that indicates sequential face arrays are preferable to simultaneous presentations [9–13]. Sequential face arrays were found to dramatically reduce the number of false identifications whilst not greatly affecting the rate of correct identifications [9–13]. This suggests that the high numbers of false hits in other studies (see e.g. [1]) using simultaneous testing procedures may be a result of the protocols employed not due to recognitions of the facial approximations alone. Therefore, the primary weakness of the facial approximation method seems to be a lack of frequent correct recognitions, rather than a vast number of false recognitions. These findings suggest that sequential face array procedures should be employed in ...

Context 2

... the simultaneous face array, facial approximation not present scenario; and (iii) the simultaneous face array, facial approximation present scenario. Thus, when facial approximations were presented to assessors in the facial approximation present scenario, the number of false identifications decreased in comparison to all other testing conditions. The similarity ratings of the facial approximation to the target individual clearly indicate high resemblance, as the modes observed for the facial approximations were high (7 out of 10 for the facial approximation with hair; and 6/10 and 8/10 for facial approximations without hair; Fig. 4). Such resemblance ratings seem comparable with reports of resemblance ratings for other ‘‘successful’’ facial approximations [14–17]. However, despite attempts to encourage ‘‘favourable’’ recognition results here (by using the facial approximation that achieved the highest resemblance, and not selecting a face pool that included distractors who looked like the target individual) recognition frequencies of the target individual were poor, and in fact tended to be less than those when assessors made an identification without the use of the facial approximation . Furthermore, when assessors used the facial approximation in sequential trials the ‘‘no identification responses’’ increased dramatically in contrast to responses when facial approximations were not used. These results offer strong support for claims that resemblance ratings do not indicate the recognizability, and hence accuracy, of facial approximations [2], and that facial approximations are not recognized frequently or reliably above rates expected by chance [1]. Results were also consistent with eyewitness research that indicates sequential face arrays are preferable to simultaneous presentations [9–13]. Sequential face arrays were found to dramatically reduce the number of false identifications whilst not greatly affecting the rate of correct identifications [9–13]. This suggests that the high numbers of false hits in other studies (see e.g. [1]) using simultaneous testing procedures may be a result of the protocols employed not due to recognitions of the facial approximations alone. Therefore, the primary weakness of the facial approximation method seems to be a lack of frequent correct recognitions, rather than a vast number of false recognitions. These findings suggest that sequential face array procedures should be employed in future facial approximation research. The lack of identification of distractor face numbers 5, 8 and 10 in recognition tests that included the facial approximation indicates that these distractor faces were not functional. That is, these faces were so dissimilar to the facial approximation that they were essentially ignored by assessors. This could be interpreted as meaning that facial approximation methods used here successfully eliminated 30% of the sample, which may perhaps be useful in forensic casework. However, we suspect that such elimination of people to whom the skull does not belong does not require the specific construction of a facial approximation, but rather it can be achieved by the use of a general description of the person’s physical characteristics as evident from the skull (e.g., broadness of face, etc.). Consequently, the lack of the identification of face numbers 5, 8, and 10 is probably better interpreted as resulting from type II bias in the face array. After all, the logic for constructing the facial approximation in the first instance is that the visual representation of the face should enable finer discrimination between individuals in contrast to a general written or verbal description. This bias in the face array acts to increase the chance level of correct identifications from 10% to 14%, suggesting that the success of the facial approximation method as observed with 10 faces is actually much less (and should only be considered with respect to 7 faces in the array). Although the large number of correct identifications made by assessors without ever seeing the facial approximation may be surprising, this can be rationally explained. As indicated in Section 1, many antemortem images obtained of target individuals are of poor resolution as these images have often been enlarged from amateur photographs. The general quality and resolution of the image of the target individual used in this study was rather poor, and this was the reason why other distractor images were resampled down. Despite these attempts it was difficult to get the pictures exactly the same even though all images were printed in newspapers. This can be attributed to the professional shooting of distractor faces (appropriate zoom, lighting, etc.) in contrast to the probable amateur shooting of the target individual under less optimal conditions. Although face numbers 2 and 10 were successfully made to be of poor quality, the target face image (face number 4) still stands out as being different (Fig. 4). We suspect this, combined with each assessor’s preconceptions of how a murder victim can be identified from a photograph, enabled a number of assessors to correctly determine who the target individual actually was. Assessor familiarity with the case or individuals in the face array can be discounted as a possible factor because assessors were requested to indicate if they recognized any of the faces either personally or as being seen in the media during the experimental trials. Two individuals tested for the face array facial approximation not present scenario, and one individual in the face array facial approximation present scenario reported familiarity with one or more individuals in the face array and were excluded from analyses reported here. These findings stress the need for any face array used for facial approximation testing to be sub- jected to tests for type I bias as slight variations in image capture of target individuals in contrast to the distractor faces (including orientation, lighting, pose, expression, etc.) may invalidate facial approximation tests. Although this study adds further support to findings that resemblance ratings do not indicate the recognizability, and hence accuracy, of facial approximations [2] and that facial approximations are rarely recognized above chance rates [1], these findings do not indicate , that ‘‘facial approximation would be detrimental to any forensic identification case’’ ([22], p. 12, [31], p. 112), or that facial approximations will never yield successful recognitions in any instance. These points need to be highlighted because they have been confused in the literature on a number of previous occasions. Identifications made from facial approximations in some instances by single individuals may result from specific and purposeful recognition by an individual, even if the overall summative rates of recognition in identification tests across individuals are not above chance levels. However, the low frequency of correct identifications suggests that the ability of facial approximations to provoke reliable recognition responses is not strong. If the facial approximation studied here generated correct recognition responses similar to that observed in other identification scenarios when the same individual’s face is used, then correct recognitions should have accounted for close to 70% of all recognitions (see e.g. [32]). Recognition responses approximating this value, or even about 50% for example, should have been easily observable within our sample if they actually existed—but they did not. The findings in this study that the recognition rate of the target individual was less than 50% and not different from chance rates at statistically significant levels, do not indicate that these recognition rates do not absolutely differ from chance rates. With very large samples recognition rates that differ very little from chance may be found to be statistically significant, however, the value of such small rates must be rationally evaluated. If facial approximations do not produce recognition rates above 50%, little confidence can be placed in identifications made by single random individuals. In forensic cases such scenarios are frequent, often with just one or a few individuals coming forward with identification statements, so if facial approximation methods are to offer some degree of reliability they must generate correct recognition responses more than 50% of the time. Until then, facial approximation methods can only be regarded as being unreliable and inaccurate. Also, while correct recognition rates of facial approximations are found to infrequently differ from chance rates, facial approximations cannot be said to provoke purposeful and specific recognitions. Of course, successful recognitions may be more frequent if other information is advertised along with the facial approximation in forensic cases, but then success is not solely dependent upon, and may have little to do with, recognition of the face that has been built on the skull [1,2,33–35]. Whilst current facial approximation methods are widely recognized as ‘‘last resort techniques’’, the abilities of the methods often appear to be much overstated [35–38]. This may contribute to decreased facial approximation success because judges have no realistic idea of what criteria they should be basing their decisions on. It may be that people familiar with facial approximation techniques (i.e., who know their strengths and weaknesses) can identify the constructed faces more accurately than lay individuals for they know what to look for and what to use in their identification decisions. For example, if it is known that the shape of the vermillion borders of the lips or the shape of the nose cannot be predicted well, but that the general position of features such as mouth location over teeth and ratio of face height to width can be well predicted, then it would seem rational not to use shape of ...

Context 3

... and included in the face array. All images were resampled 1 down (pixels removed), scaled and cropped in Adobe 1 Photoshop 6.0 to give images as closely comparable as possible to each other (Fig. 3). No effort was made to select individuals of similar physical appearance (in this case facial appearance) to the target as recommended in usual eyewitness tests. Thus, this face array consisted of a fairly random sample and is expected to provide a scenario favourable for correct recognitions of the facial approximation (the face array may be biased and hence some distractor faces may not be seen to be plausible alternatives to the target individual and/or the facial approximation). To determine if any type I bias was present in the face array, sequential and simultaneous presentation trials were conducted without the facial approximation. Fifteen adult assessors (11 females, 4 males: mean age 23 years, standard deviation 12 years) who did not recognize, either personally or as having been seen in the media, any of the faces in the face array were asked to determine, without the facial approximation, who the murder victim (target individual) was in the face array. Assessors had no information apart from the photographs themselves to base this decision on. Assessors first participated in the sequential face array presentation and then the simultaneous line-up. In the sequential face array assessors where shown one face image at a time in a random order. Assessors were forced to decide for each face ‘‘yes’’ or ‘‘no’’ if the face was that of the murder victim. Assessors were aware that if they made an identification of ‘‘yes’’ the trial was completed (they would not see the other faces in the sequence during this test) and if they answered ‘‘no’’ that they would not be able to change their mind to this face at a later point in the trial. The investigator (CNS) held more cards than those included in the face array test so that assessors could not anticipate the end of the sequence. In the sequential trial cards were held about 1 m in front of the assessor and at arm’s length from the investigator at 90% to his line of sight so investigator cues, if there were any, were not obvious to assessors (who were hopefully concentrating on the cards and hence not attending to any cues, particularly facial ones, if expressed by the investigator). After assessors had made an identification, or if they proceeded through all the faces in the sequence without making an identification, all the faces in the array were presented simultaneously, but in a random order, to the assessor for him/her to change their identification decision from the sequential trial if they so wished. Twenty new assessors (14 females, 6 males: mean age 20 years, standard deviation 4 years) who did not recognize, either personally or as having been seen in the media, any of the faces in the face array, were recruited for the main project: to attempt to correctly identify who the target individual was from the facial approximation. Identical procedures were followed as indicated above for face pool testing without the facial approximation, except of course that this time the assessors had access to a facial approximation. Although assessors had the option of not choosing any face in the simultaneous line-up, a chance rate of 10% was used as it was found that almost all assessors selected a face (see Section 3), and hence appeared biased in this respect (according to chance one would expect a 50% response of ‘‘not there’’, instead it seemed all assessors were choosing a face and hence each face in the array had a 10% chance of being selected). Responses were recorded categorically as correct (target face identification) or incorrect (distractor identification or ‘‘no identification’’ response) for statistical analysis. Observed data were compared to expected frequen- 1 cies by Fisher’s Exact tests in the JMP 4.0 statistical package. As we were only interested in responses larger than chance (and because chance rates were so small that lesser differences would be difficult to detect) confidence intervals for one tailed tests were used. Assessor’s resemblance ratings of both facial approximations were in accordance with previous indications from other individuals, including other forensic facial approximation experts, that resemblance was high. It is worth noting that Betty Pat Gatliff, a high profile forensic artist and facial approximation practitioner, indicated to the first author in 2000 that the face appeared similar enough that she would have expected a positive result had it been advertised. Both facial approximations received high, but similar, resemblance rating results (around 7 out of 10), although the facial approximation without hair tended to be rated higher than the facial approximation with hair (Fig. 4). Data distributions for the no hair facial approximation displayed slightly more left skew than those for the facial approximation with hair (Fig. 4). Hence in recognition tests reported here we used the facial approximation without hair as this is expected to favour positive recognition responses according to traditional facial approximation theory (greater resemblance, greater accuracy and possibly more frequent correct recognitions). Sequential and simultaneous face array tests, done without assessors seeing a facial approximation, showed that the target individual (face number 4) was identified at rates well above other individuals (Fig. 5) and that responses were highly similar between the sequential and simultaneous presentation methods. For sequential line-ups, the identification of the target individual (face number 4) in comparison to other faces was statistically significant ( p < 0.08) in six out of nine cases (i.e., for face numbers 1, 2, 5, 6, 7, and 8), although the other three cases (face numbers 3, 9, and 10) followed similar trends. In no case was a ‘‘not there’’ response made for either the sequential or the simultaneous face array. During post-experiment feed back assessors expressed their thoughts why they thought they could tell who the murder victim was without the facial approximation. Frequently these explanations included: ‘‘the target photograph . . . would not be flattering’’, ‘‘ . . . would not be clear’’, and/or ‘‘ . . . would present the person smiling’’. When the facial approximation without hair was presented to assessors who then attempted to identify the target individual from the sequential face array, identification of the target individual (face number 4) did not increase (Fig. 6). Face number 4 was not identified above chance rates at statistically significant levels. Face number 4 was also identified some- what less, but not at rates statistically different from that observed when the facial approximation was not presented to assessors (in both sequential and simultaneous procedures). During sequential trials the majority of responses were the default response of ‘‘no identification’’, as most individuals (60%) completed the sequential presentation without identifying any face. The increased number of ‘‘not there’’ responses for the sequential face array, facial approximation present scenario, was highly statistically significant ( p < 0.001) in comparison to: (i) the sequential face array, facial approximation not present scenario; (ii) the simultaneous face array, facial approximation not present scenario; and (iii) the simultaneous face array, facial approximation present scenario. Thus, when facial approximations were presented to assessors in the facial approximation present scenario, the number of false identifications decreased in comparison to all other testing conditions. The similarity ratings of the facial approximation to the target individual clearly indicate high resemblance, as the modes observed for the facial approximations were high (7 out of 10 for the facial approximation with hair; and 6/10 and 8/10 for facial approximations without hair; Fig. 4). Such resemblance ratings seem comparable with reports of resemblance ratings for other ‘‘successful’’ facial approximations [14–17]. However, despite attempts to encourage ‘‘favourable’’ recognition results here (by using the facial approximation that achieved the highest resemblance, and not selecting a face pool that included distractors who looked like the target individual) recognition frequencies of the target individual were poor, and in fact tended to be less than those when assessors made an identification without the use of the facial approximation . Furthermore, when assessors used the facial approximation in sequential trials the ‘‘no identification responses’’ increased dramatically in contrast to responses when facial approximations were not used. These results offer strong support for claims that resemblance ratings do not indicate the recognizability, and hence accuracy, of facial approximations [2], and that facial approximations are not recognized frequently or reliably above rates expected by chance [1]. Results were also consistent with eyewitness research that indicates sequential face arrays are preferable to simultaneous presentations [9–13]. Sequential face arrays were found to dramatically reduce the number of false identifications whilst not greatly affecting the rate of correct identifications [9–13]. This suggests that the high numbers of false hits in other studies (see e.g. [1]) using simultaneous testing procedures may be a result of the protocols employed not due to recognitions of the facial approximations alone. Therefore, the primary weakness of the facial approximation method seems to be a lack of frequent correct recognitions, rather than a vast number of false recognitions. These findings suggest that sequential face array procedures should be employed in future facial approximation research. The lack of identification of distractor face numbers 5, 8 and 10 in recognition tests that included the facial ...

Context 4

... facial approximation. Identical procedures were followed as indicated above for face pool testing without the facial approximation, except of course that this time the assessors had access to a facial approximation. Although assessors had the option of not choosing any face in the simultaneous line-up, a chance rate of 10% was used as it was found that almost all assessors selected a face (see Section 3), and hence appeared biased in this respect (according to chance one would expect a 50% response of ‘‘not there’’, instead it seemed all assessors were choosing a face and hence each face in the array had a 10% chance of being selected). Responses were recorded categorically as correct (target face identification) or incorrect (distractor identification or ‘‘no identification’’ response) for statistical analysis. Observed data were compared to expected frequen- 1 cies by Fisher’s Exact tests in the JMP 4.0 statistical package. As we were only interested in responses larger than chance (and because chance rates were so small that lesser differences would be difficult to detect) confidence intervals for one tailed tests were used. Assessor’s resemblance ratings of both facial approximations were in accordance with previous indications from other individuals, including other forensic facial approximation experts, that resemblance was high. It is worth noting that Betty Pat Gatliff, a high profile forensic artist and facial approximation practitioner, indicated to the first author in 2000 that the face appeared similar enough that she would have expected a positive result had it been advertised. Both facial approximations received high, but similar, resemblance rating results (around 7 out of 10), although the facial approximation without hair tended to be rated higher than the facial approximation with hair (Fig. 4). Data distributions for the no hair facial approximation displayed slightly more left skew than those for the facial approximation with hair (Fig. 4). Hence in recognition tests reported here we used the facial approximation without hair as this is expected to favour positive recognition responses according to traditional facial approximation theory (greater resemblance, greater accuracy and possibly more frequent correct recognitions). Sequential and simultaneous face array tests, done without assessors seeing a facial approximation, showed that the target individual (face number 4) was identified at rates well above other individuals (Fig. 5) and that responses were highly similar between the sequential and simultaneous presentation methods. For sequential line-ups, the identification of the target individual (face number 4) in comparison to other faces was statistically significant ( p < 0.08) in six out of nine cases (i.e., for face numbers 1, 2, 5, 6, 7, and 8), although the other three cases (face numbers 3, 9, and 10) followed similar trends. In no case was a ‘‘not there’’ response made for either the sequential or the simultaneous face array. During post-experiment feed back assessors expressed their thoughts why they thought they could tell who the murder victim was without the facial approximation. Frequently these explanations included: ‘‘the target photograph . . . would not be flattering’’, ‘‘ . . . would not be clear’’, and/or ‘‘ . . . would present the person smiling’’. When the facial approximation without hair was presented to assessors who then attempted to identify the target individual from the sequential face array, identification of the target individual (face number 4) did not increase (Fig. 6). Face number 4 was not identified above chance rates at statistically significant levels. Face number 4 was also identified some- what less, but not at rates statistically different from that observed when the facial approximation was not presented to assessors (in both sequential and simultaneous procedures). During sequential trials the majority of responses were the default response of ‘‘no identification’’, as most individuals (60%) completed the sequential presentation without identifying any face. The increased number of ‘‘not there’’ responses for the sequential face array, facial approximation present scenario, was highly statistically significant ( p < 0.001) in comparison to: (i) the sequential face array, facial approximation not present scenario; (ii) the simultaneous face array, facial approximation not present scenario; and (iii) the simultaneous face array, facial approximation present scenario. Thus, when facial approximations were presented to assessors in the facial approximation present scenario, the number of false identifications decreased in comparison to all other testing conditions. The similarity ratings of the facial approximation to the target individual clearly indicate high resemblance, as the modes observed for the facial approximations were high (7 out of 10 for the facial approximation with hair; and 6/10 and 8/10 for facial approximations without hair; Fig. 4). Such resemblance ratings seem comparable with reports of resemblance ratings for other ‘‘successful’’ facial approximations [14–17]. However, despite attempts to encourage ‘‘favourable’’ recognition results here (by using the facial approximation that achieved the highest resemblance, and not selecting a face pool that included distractors who looked like the target individual) recognition frequencies of the target individual were poor, and in fact tended to be less than those when assessors made an identification without the use of the facial approximation . Furthermore, when assessors used the facial approximation in sequential trials the ‘‘no identification responses’’ increased dramatically in contrast to responses when facial approximations were not used. These results offer strong support for claims that resemblance ratings do not indicate the recognizability, and hence accuracy, of facial approximations [2], and that facial approximations are not recognized frequently or reliably above rates expected by chance [1]. Results were also consistent with eyewitness research that indicates sequential face arrays are preferable to simultaneous presentations [9–13]. Sequential face arrays were found to dramatically reduce the number of false identifications whilst not greatly affecting the rate of correct identifications [9–13]. This suggests that the high numbers of false hits in other studies (see e.g. [1]) using simultaneous testing procedures may be a result of the protocols employed not due to recognitions of the facial approximations alone. Therefore, the primary weakness of the facial approximation method seems to be a lack of frequent correct recognitions, rather than a vast number of false recognitions. These findings suggest that sequential face array procedures should be employed in future facial approximation research. The lack of identification of distractor face numbers 5, 8 and 10 in recognition tests that included the facial approximation indicates that these distractor faces were not functional. That is, these faces were so dissimilar to the facial approximation that they were essentially ignored by assessors. This could be interpreted as meaning that facial approximation methods used here successfully eliminated 30% of the sample, which may perhaps be useful in forensic casework. However, we suspect that such elimination of people to whom the skull does not belong does not require the specific construction of a facial approximation, but rather it can be achieved by the use of a general description of the person’s physical characteristics as evident from the skull (e.g., broadness of face, etc.). Consequently, the lack of the identification of face numbers 5, 8, and 10 is probably better interpreted as resulting from type II bias in the face array. After all, the logic for constructing the facial approximation in the first instance is that the visual representation of the face should enable finer discrimination between individuals in contrast to a general written or verbal description. This bias in the face array acts to increase the chance level of correct identifications from 10% to 14%, suggesting that the success of the facial approximation method as observed with 10 faces is actually much less (and should only be considered with respect to 7 faces in the array). Although the large number of correct identifications made by assessors without ever seeing the facial approximation may be surprising, this can be rationally explained. As indicated in Section 1, many antemortem images obtained of target individuals are of poor resolution as these images have often been enlarged from amateur photographs. The general quality and resolution of the image of the target individual used in this study was rather poor, and this was the reason why other distractor images were resampled down. Despite these attempts it was difficult to get the pictures exactly the same even though all images were printed in newspapers. This can be attributed to the professional shooting of distractor faces (appropriate zoom, lighting, etc.) in contrast to the probable amateur shooting of the target individual under less optimal conditions. Although face numbers 2 and 10 were successfully made to be of poor quality, the target face image (face number 4) still stands out as being different (Fig. 4). We suspect this, combined with each assessor’s preconceptions of how a murder victim can be identified from a photograph, enabled a number of assessors to correctly determine who the target individual actually was. Assessor familiarity with the case or individuals in the face array can be discounted as a possible factor because assessors were requested to indicate if they recognized any of the faces either personally or as being seen in the media during the experimental trials. Two individuals tested for the face array facial approximation not present scenario, and one individual in the face array facial ...