Collocations are a ubiquitous phenomenon. Without consciously thinking about them, people encounter them continuously during language use in both comprehension and production. In linguistics, however, collocations have been extensively studied in the last one hundred years and often characterized either as phraseological phenomena (e.g., bitter enemy) or frequent co-occurrences (e.g., annual meeting). In particular, the two structures that were paid most attention to are verb+noun (e.g., abuse trust) and adjective+noun (e.g., high hopes). And although there is evidence from various fields (e.g., first language acquisition, corpus linguistics, cognitive grammar to name a few) about selection preferences towards the use of certain categories in general and for collocations in particular, there are still no comprehensive studies of collocations involving a broad range of different structural groups. Therefore, one of the aims of this thesis is to study whether there are any preferences in the collocational competence towards certain structural groups and if yes, whether this preference is different across languages. For this reason, over 200 native and advanced non-native speakers of English and German as well as native speakers of Russian were tested on their knowledge of phraseological collocations using a multiple-choice collocations test in Experiment 1. Within the same experiment, it was attempted to discover the predictors of collocation competence: e.g., reading habit, alphabetic script, education level, age, onset of language learning, residence in the L2 country, etc. It was established that the morpho-syntactic properties of the language as well as the phenomenon of collocation predict collocation competence. For example, participants were significantly better in those collocational structures the components of which were adjacent and were not often interrupted by insertions in forms of other lexical units. In addition, given the fact that collocational competence can be facilitated by reading, the reading habit/skill was measured using three different measures though various questions in the paper-based questionnaire and in Experiment 2 which was a self-paced reading task. Surprisingly, it was discovered that certain measures of the reading habit can predict collocations reliably better than others. For example, asking a participant how much time they spend reading per week significantly predicted collocational competence in all native speaker groups (i.e., English, German and Russian), but not the non-native groups. The ii contribution of this finding could be methodologically valuable for researchers searching for correlations between the reading habit and different aspects of language use. According to another interesting finding, the similarity of the orthographic systems of L1 and L2 is associated with better collocational competence. In other words, the collocational competence of non-native speakers of German having alphabetic script similar to the L2 (e.g., Ukrainian, Hungarian or Greek) was more likely to be better than of those learners whose alphabetic script was different from the L2 (e.g., Chinese, Arabic, Armenian) suggesting that similarity of script would result in an easier recognition of patterns in the L2 if the script of the L2 is familiar to the language users. It was then hypothesized that this selection preference towards particular collocational structures might stretch to processing differences. Thus, the same participants were tested in Experiment 3, which was a phrasal decision task, and which revealed that adjective+noun collocations were processed the fastest among the four structural groups of collocations. The results also showed that phrasal-like collocations (i.e., adjective+noun and adverb+adjective) were generally processed faster than clausal-like collocations (i.e., verb+noun and verb+adverb) suggesting that processing collocations is easier and faster when the argument structure construction involving the verb does not have to be activated. Processing collocations of various structures was also dependent on whether the base of the collocation belonged to the primary categories from cognitive grammar or to secondary ones. Thus, collocation containing bases that belonged to primary categories (e.g., things or actions) were processed significantly faster than collocations containing words from secondary categories (e.g., qualities). However, the effect is not robust across the five groups tested and it remains unclear whether this effect is specific to the processing of collocations, or it unfolds on higher levels of language processing. These results suggest that a theory of collocation processing should be explicit about the base of the collocation and what role it plays in the processing of a collocation. Similarly to previous findings, it was also established that collocations are processed faster than free combinations suggesting that collocations as chunks (e.g., bitter enemy) are processed in a holistic fashion in comparison to free combinations (e.g., common enemy) that are novel language. The effect was robust across all participant groups, both native and non-native, which speaks in favour of the fact that collocations are a psychologically valid phenomenon for both native and non-native populations. iii Finally, it has been attempted to study the neural correlates of collocation processing. In Experiment 4, available (electroencephalography) EEG measurements from 31 participants listening to a German audio book were analyzed. Given the fact that the speech stimuli were pre-determined, i.e., the text of the audio book, the interpretation of collocation in this part of the study was different from that employed in Experiments 1- 3. Therefore, collocations were seen as frequently co-occurring word combinations which are different from non-collocations based on a corpus-derived association measure, in this case, mutual information. The brain response for the two conditions was compared in the N400 time window. It was hypothesized that since N400 is an ERP component that is a marker of unpredictability, surprisal and higher processing load, non-collocations should exhibit a larger N400 in comparison to collocations. The results of the permutation tests revealed that collocations were processed statistically differently from non-collocations at all electrode sites. The exact configurations of the effect, however, are mixed. The largest differences were observed in the left-anterior area where collocations were modulated by larger negativities in the N400 time window, and in the right-posterior area where non-collocations displayed larger negativities. Given that previous studies of collocations (including Experiment 3 of the current thesis) often employed artificial designs in studying these word combinations, the result of Experiment 4 are especially valuable for the understanding of collocations since the target items were neither matched with non-collocations containing semantic violation nor presented in isolation. In fact, all items were embedded in a coherent piece of discourse presented in the form of a continuous speech signal that presents a highly naturalistic task. To sum up, this work has managed to provide evidence that collocations are both psychologically and neurophysiologically valid using both standard behavioural paradigms as well as elaborate electrophysiological (naturalistic) methods, which presents triangulation of methods in the study of collocation processing. Given that the differences between collocation processing in native and non-native populations have been extensively studied previously, a further step can be seen in researching the neural correlates of collocation processing in non-native speakers, which will bring us closer to gaining a comprehensive picture of the nature of collocations.