When recognition probes seem familiar but their presentation cannot be recollected, dual-process models predict that they will be attributed to too many presentation contexts--most dramatically, to multiple contexts that are mutually contradictory. This is the phenomenon of episodic over-distribution. In the conjoint-recognition and process-dissociation paradigms, attributions to two contradictory contexts can be measured: (a) presented and not presented (conjoint recognition) and (b) presented on List 1 only and presented on List 2 only. Consistent with dual-process models but inconsistent with one-process models, analyses of over 100 sets of conjoint-recognition data revealed that attribution of probes to the first contradictory combination was virtually universal. Across the data corpus, 18% of true-memory probes (studied targets) and 13% of false-memory probes (related distractors) were judged to have been both presented and not presented. Likewise, episodic over-distribution was detected in follow-up analyses of process-dissociation data sets, where an average of 39% of target probes were judged to have been presented on List 1 only and to have been presented on List 2 only.