Cox & Shiffrin (2012) introduced a dynamic model of recognition memory that was capable of making simultaneous predictions for accuracy and mean response time. The present paper extends that work by investigating the assumptions underlying the model's decision process, in particular those pertaining to the process by which features are sampled at test and the processes by which evidence for an "old"/"new" recognition decision accumulates. These assumptions are tested against empirically collected response time distributions. Evidence is found that sampling dynamics can change in response to instructions, and that both independent and correlated accumulators for "old" and "new" evidence are viable mechanisms for explaining accuracy and response time data.