Retention was assessed using a four-alternative recognition test in a modified Brown-Peterson paradigm. Performance decreased with the length of the distraction interval at a faster rate when the test modality (auditory or visual) did not match the presentation modality than when test and presentation modalities did match. These results, which were replicated in a second experiment, were interpreted in terms of a dual-access model of the recognition process and a feature conception of memory codes. Also, in Experiment 1, modality-specific encoding was not circumvented by dual-modality presentation (auditory plus visual): dual-modality presentation resulted in performance comparable to that observed following visual presentation. When subjects were instructed to attend to both modalities equally in Experiment 2, the pattern of results reflected a corresponding shift in attentional bias but modality-dependent encoding was circumvented only partially. This result was interpreted as being inconsistent with either the notion of modality-specific processing capacities or dual-code theory.