Three experiments are presented that investigated the recognition of words after their acoustic offsets in conversational speech. Utterances randomly selected from the speech of24 individuals (total N=288) were gated in one-word increments and heard by 12 listeners each. Of the successful recognitions, 21 %occurred after the acoustic offset of the word in question and in the presence of subsequent context. The majority of late recognitions implicate subsequent context in the recognition process. Late recognitions were distributed nonrandomly with respect to the characteristics of the stimulus word tokens. Control experiments demonstrated that late recognitions were not artifacts of eliminating discourse context, of imposing artificial word boundaries, or of repeating words within successive gated presentations. The effects could be replicated only if subsequent context was available. The implications are discussed for models of word recognition in continuous speech.A decade and a half of research on the recognition of words in fluent speech has demonstrated that under the right conditions listeners can apply acoustic and higherlevel information to this task more or less instantaneously (Grosjean, 1980;Marslen-Wilson, 1973;Marslen-Wilson & Tyler, 1980;Marslen-Wilson & Welsh, 1978). Models of word recognition designed around these fmdings have consequently emphasized the optimal functioning of word recognition processes. The following descriptions of the process of listening to speech appear to describe a device that utilizes preceding context only and that starts a new cycle at the onset of each new word:Utterances are indeed understood as they are heard; ... the listenerconstructsa syntactic and semanticinterpretation of the input word-by-word as he hears it, and ... he actively uses this information to guide his processing of the subsequent wordsin the string. (Marslen-Wilson, Tyler, & Seidenberg, 1978, p. 240) Speech is processed sequentially, word by word ... the words in an utterance are recognized one after another ... listenersknow where wordsbeginand end by recognizing them in order. (Cole & Jakimik, 1980, pp. 133-134) We suggestthat in perception of conversational speechunder good listening conditions, recognition of one word is This study was supported by Grant GRlC78377 from the Science and Engineering Research Council (U.K.) to H. S. Thompson and E. G. Bard. Speech materials were drawn from those prepared under the auspices of grants by the Education and Social Science Research Council (U.K.) to J. Laver and E. G. Bard (HR 6130) and to G. Brown (HR 3601 complete before recognition of the following word begins. (Cole & Jakimik, 1980, p. 149) Increasingly, however, it has been suggested that the human speech recognition mechanism operates in a more varied fashion than these statements imply. Experimental evidence of less than optimal processing dates from the experiments of Pickett and Pollack (1963;, 1964, who used a gating technique in which listeners were asked to identify all the words con...