All second language (L2) learning theories presuppose that learners learn the target language from the speech signal (or written material, when learners are reading), so an understanding of learners' ability to detect and represent novel patterns in linguistic stimuli will constitute a major building block in an adequate theory of second language acquisition (SLA) input. Pattern detection, a mainstay of current connectionist modeling of language learning, presupposes a sensitivity to particular properties of the signal. Learning abstract grammatical knowledge from the signal presupposes, as well, the capacity to map phonetic properties of the signal onto properties of another type (segments and syllables, morpheme categories, and so on). Thus, even seemingly "simple" grammatical phenomena may embody complex structural knowledge and be instantiated by a plethora of diverse cues. Moreover, cues have no a priori status; a phenomenon of a given sort takes on a value as a cue when acquisition of the grammatical system reveals it to be useful. My study deals with initial sensitivity to cues to gender attribution in French. Andersen (1984) asked: "What's gender good for anyway?" One answer comes from a number of studies, done mostly in the last 20 years, of gender processing by both monolingual and bilingual speakers (among many others, Bates, Devescovi, Hernandez, & Pizzamiglio, 1996; Bates & Liu, 1997; Friederici & Jacobsen, 1990; Grosjean, Dommergues, Cornu, Guillemon, & Besson, 1994; Guillemon & Grosjean, 2001; Taft & Meunier, 1998). These studies provide evidence that in monolinguals and early (but not late) L2 learners, prenominal morphosyntactic exponents of gender prime noun activation and speed up noun recognition. Over the same period, a growing number of studies detailing the course of L2 gender acquisition for a variety of different target languages and learner types (e.g., Bartning, 2000; Chini, 1995; Dewaele & Veronique, 2000; Granfeldt, 2003; Hawkins & Franceschina, 2004) have provided support for the hypothesis that developmental paths differ for early and later learners of gender. Yet despite its obvious importance to SLA theorizing, few studies have dealt directly with adult learners' ability to detect and analyze potential cues to gender at the initial stage of exposure to the L2 (and this despite considerable discussion in recent years of the nature of the "initial state" of L2 learning). The study reported on in this article, which was actually conducted in the late 1980s, was an attempt to shed some light on what the beginning learner can do with the gender attribution problem. This study was, at that time, and is even now, an anomaly; most research dealing with "input" provided descriptions of what people say to learners, not what learners can perceive and represent. Indeed, most studies that shed light on the initial analytical capacities of absolute beginners were concerned with "perceptual" learning, that is, with the acquisition of phonetic or phonological distinctions (e.g., Broselow, Hurtig, & Ringen's [19871 study of tone learning or various studies on the perception of the /r/ vs. /l/ phonemes in American English by Japanese speakers). In this update, it is therefore worth mentioning Rast's (2003) dissertation and Rast and Dommergues (2003), which is based on it, which examined the results of the first 8 hr of instructed learning of Polish by francophone adults. My study asked if anglophone adults, with little or no prior exposure to French, given auditory stimuli, were equally sensitive to phonological, morphosyntactic, or semantic cues to French gender classes. The issue of what learners can detect in the signal and encode is an empirical one. I presented 88 adult English speakers with highly patterned data in list form, namely, auditory sequences of [Det+N](French)+translation equivalent(English) forms. The patterns, all true generalizations, were drawn from linguistic descriptions of French. These cues are believed by grammarians of the language to be "psychologically real" to native speakers. I then measured in 3 different ways what my participants had acquired. Given the extreme limitations on the input (no visual supports to identify referents of names), the participants performed pretty well. Moreover, they proved to be highly sensitive to "natural" semantic and morphological patterns and could generalize accurately from learned instances to novel exemplars. These patterns, however, are not directly instantiated in the speech signal; they are abstractions imposed on the stimuli by human linguistic cognition. Moreover, although it would be inaccurate to describe the learning patterns as "transfer" (because English nouns have no gender feature), prior knowledge seemed to be implicated in the results. Above all, these Anglophones appear to perceive the gender learning problem as a semantic one and to make use of "top-down" information in solving it. It follows that the pattern detection that they can do when listening to speech is clearly biased by what they already know. These results, therefore, provide support for hypotheses that the initial state is to be defined in terms of the transfer of first language (L1) grammatical knowledge and/or the transfer of L1-based processing procedures.