Event-related fMRI was employed to investigate the influence of the relative probability of old and new test items on the neural correlates of recognition memory. Twelve subjects undertook three study-test cycles, each consisting of an identical study phase in which a series of words was encoded in an incidental task, followed by a test phase in which yes/no recognition judgments were made to a mixture of studied (old) and unstudied (new) words. The ratio of old to new words differed in each test phase, and was either 25:75, 50:50, or 75:25. In lateral inferior and medial parietal cortex, and the posterior cingulate, greater activity was elicited by correctly classified old than new items independently of old:new ratio. By contrast, in other regions, including anterior, dorsolateral, and ventrolateral prefrontal cortex, differences in the activity elicited by old and new items varied according to old:new ratio, demonstrating in some cases a complete crossover interaction. The results suggest that differential activity elicited by old and new test items is likely to support successful recognition in only a subset of the regions identified in previous studies as exhibiting such differences. In other regions, most notably prefrontal cortex, differences in the activity elicited by old and new items appear to reflect processes that are contingent upon, rather than in support of, successful recognition.