We describe here four experiments which deal with the contrast between two methods of investigating memory for prior episodes: perceptual identification and recognition. This work extends earlier models of word identification following word lists and single sentences to identification and recognition of words following processing of naturally occurring discourse. In experiment 1 subjects read five passages containing typical and atypical actions within an everyday script; subjects were tested immediately for recognition of words selected from typical and atypical actions. Discriminability was better for atypical words that had been seen than for typical words. In experiment 2 identification of the same words exposed for brief duration was required. Words related to both typical and atypical events were identified equally well and both more accurately than new unrelated words. In experiment 3 typical, atypical, and unrelated words were interspersed in the identification list with words related to actions directly inferrable from typical and atypical events in the stories, but which never appeared in the story. Typical words were identified more accurately than atypical words. However, inferrable words which related to typical events were recognised at an equal level with atypical old words, and all inferrable words were superior to unrelated words. A similar pattern obtained in the correct response times for the different groups of words. In experiment 4 subjects identified the five categories of words above: (a) when only a title but no story had been provided, and (b) when neither title nor story was provided. No differences in perceptual accuracy were found in either condition. These results support models of word identification in which both featural and temporal-contextual elements of an item play a part in both perception and recognition. �� 1987.