Models of recognition memory have traditionally struggled with the puzzle of criterion setting, a problem that is particularly acute in cases in which items for study and test are of widely varying types, with differing degrees of baseline familiarity and experience (e.g., words vs. random dot patterns). We present a dynamic model of the recognition process that addresses the criterion setting problem and produces joint predictions for choice and reaction time. In this model, recognition decisions are based not on the absolute value of familiarity, but on how familiarity changes over time as features are sampled from the test item. Decisions are the outcome of a race between two parallel accumulators: one that accumulates positive changes in familiarity (leading to an ‘‘old’’ decision) and another that accumulates negative changes (leading to a ‘‘new’’ decision). Simulations with this model make realistic predictions for recognition performance and latency regardless of the baseline familiarity of study and test items.