Associative recognition has been attributed to the operation of a recall-like process that is not engaged during recognition of single items. An alternative mechanism is the formation of a compound memory cue that incorporates relational information between the elements of the association. These accounts make different predictions about the dynamics of associative recognition as revealed by speed-accuracy trade-off (SAT) functions: if recall were operating, SAT functions should approach asymptotic performance at a faster rate for stronger associations. In a review of a literature, we find instead that only asymptotic performance, not rate, is affected by associative strength, consistent with a compound cue mechanism. We present a formal model of this mechanism as a direct outgrowth of a model of single-item recognition (Cox & Shiffrin, 2012) and use it to predict observed SAT curves for both single-item and associative recognition in a variety of experiments.