Summary. The study examines the possibility of observing on-line recognition of spoken words through manipulations of the location of the uniqueness point (UP) in a gender-classification task. The subjects were presented with spoken French nouns and had to indicate by a key-press response whether each 'was feminine or masculine. RTs measured from word onset were significantly correlated with UP location, a finding that supports the notion of online processing. The effect of UP location is, however, smaller than that predicted by the original cohort theory (that recognition occurs exactly at the UP). On the other hand, it is stronger when words with respectively early and late UPs are presented in homogeneous blocks rather than in mixed order. It is proposed that the results can be accounted for by the notion of a sub-optimal lexical strategy in which some monitoring of the phonetic data continues past the UP.A notion that has exerted a great influence on current conceptions of speech perception is that the process of word recognition begins on line during reception of the acoustic evidence (Marslen-Wilson & Welsh, 1978;Cole & Jakimik, 1980;Frauenfelder & Tyler, 1987). As Cole and Rudnicky (1983) note, the notion was anticipated a century ago by Bagley (1900-1). Marslen-Wilson and Welsh (1978) gave the notion a strict application in their "cohort model" in which speech recognition was considered as a real-time process making optimal use of the information in the mental lexicon. For the case of isolated words, cohort theory assumed that full identification occurs as soon as the data processed since the beginning of the word are no longer compatible with more than one lexical candidate. The point at which the critical batch of information bringing the set of candidates (the "cohort") down to one is delivered is called the "uniqueness point" (UP). For pseudowords used in the lexical decision task, the critical statistic is the "deviation point" (DP), i.e., the point at which the data cease to be compatible with any item represented in the lexicon.The existing evidence from studies of the effects of either UP or DP location is somewhat contradictory.A strong effect of the location of the UP was reported by Marslen-Wilson (1980) in a phoneme-monitoring task using lists of disconnected spoken words. The original aim Offprint requests to: M. Radeau of the study was to find out whether the task was performed at the level of retrieved lexical representations. A regular decrease in monitoring time was observed, the further the target occurred in the word. The data were subsequently (Marslen-Wilson, 1987) reanalysed in relation to the distance between the target and the UP and a very high correlation was found. The slope, however, was not reported, making it difficult to estimate the size of the effect.Luce, Pisoni, and Manous (1984), using a "gating task" in which the subject is asked to identify a word on the basis of an early segment of varying size, obtained an effect of the location of the UP. The effect was probably s...