We investigated source misattributions in the DRM false memory paradigm (Deese, 1959, Roediger & McDermott, 1995). Subjects studied words in one of two voices, manipulated between-lists (pure-voice lists) or within-list (mixed-voice lists), and were subsequently given a recognition test with voice-attribution judgements. Experiments 1 and 2 used visual tests. With pure-voice lists (Experiment 1), subjects frequently attributed related lures to the corresponding study voice, despite having the option to not respond. Further, these erroneous attributions remained high with mixed-voice lists (Experiment 2). Thus, even when their related lists were not associated with a particular voice, subjects misattributed the lures to one of the voices. Attributions for studied items were fairly accurate in both cases. Experiments 3 and 4 used auditory tests. With pure-voice lists (Experiment 3), subjects frequently attributed related lures and studied items to the corresponding study voice, regardless of the test voice. In contrast, with mixed-voice lists (Experiment 4), subjects frequently attributed related lures and studied items to the corresponding test voice, regardless of the study voice. These findings indicate that source attributions can be sensitive to voice information provided either at study or at test, even though this information is irrelevant for related lures.