Two studies examined the cognitive units of sentence memory using a perceptual recognition task. Four candidate cognitive units were considered: concepts, propositions, integrated propositions, and nonintegrated propositions. Subjects first received a list of acquisition sentences and then were asked to reproduce sentences presented under a white-noise mask. These masked sentences were replicas of the acquisition sentences, were formed of recombined clauses from the acquisition sentences, or were formed of recombined words from the acquisition sentences. Reproduction accuracy was employed as the dependent measure. Results supported propositions (operationalized by clauses) as cognitive units of episodic memory. No conclusive evidence was obtained for concepts, integrated propositions, or nonintegrated propositions as cognitive units. The utility of perceptual recognition tasks for studying the cognitive units of episodic memory is discussed. Recently, Hannigan, Shelton, Franks, & Bransford (1980) have examined the role of episodic and semantic memory factors (cf. Tulving, 1972) in facilitating the perceptual recognition of sentences masked by white noise. In this work, experimental groups were presented a list of seemingly unrelated sentences as acquisition materials. These sentences could be made to seem interrelated through knowledge of a framework that gave each of them a more specific contextual meaning. One group of subjects was given the framework during acquisition (framework group). A second group was only presented with the sentences (no-framework group). It was hypothesized that presentation of the sentences would result in encoded episodic representations of the information in the sentences. Knowledge of the framework, however, was expected to add higher order semantic knowledge structures within which episodic information would be organized.Following acquisition, the effects of these episodic and semantic factors on perceptual recognition were assessed (cf. Jacoby & Dallas, in press). The perceptual recognition task consisted of presenting sentences masked by white noise and asking subjects to reproduce the sentences. Two types of test sentences were old sentences (which were actually presented during acquisition) and novel-appropriate sentences (which were This work was supported in part by a National Science Foundation predoctoral fellowship (SMI772098l) to Pamela M. Auble. The authors wish to express their thanks to Tommie Slayden and Kate Bergmann for their help in designingand conducting the experiments. We would also like to thank Chuck Clifton, James Neely, and an anonymous reviewer for their helpful comments and criticism of an earlier draft of this paper. Requests for reprints should be sent to Jeffery J. Franks, Department of Psychology, Vanderbilt University, Nashville, Tennessee 37240. not previously presented but did fit the framework provided during acquisition). It was found that both the framework and no-framework groups were more accurate in reproducing old sentences than a contro...