Disjunction fallacies have been extensively studied in probability judgment. They should also occur in episodic memory, if remembering a cue's episodic state depends on how its state is described on a memory test (e.g., being described as a target vs. as a distractor). If memory is description-dependent, cues will be remembered as occupying logically impossible combinations of episodic states (e.g., as being a target "and" a distractor). Consistent with this idea, memory disjunction fallacies were repeatedly detected in a series of experiments, at the level of individuals as well as at the level of groups. Disjunction fallacies varied as a function of manipulations that should affect description-dependency, such as type of test cue, immediate versus delayed testing, word frequency, and emotional valence. Response bias, as well as description-dependency, contributed to disjunction fallacies, as predicted by fuzzy-trace theory's retrieval model. The significance of these findings for memory is that a new form of episodic distortion, description-dependent memory, has been added to the 2 traditional forms (forgetting and false memory). The significance for probability judgment is that disjunction fallacies, which have customarily been explained as by-products of memory retrieval, may be wholly or partly due to the uncontrolled influence of response bias. (Contains 1 figure, 4 tables and 3 footnotes.)