Recollection has long been thought to play a key role in associative recognition tasks. Evidence that associative recollection might be a threshold process has come from analyses of the associative recognition receiver operating characteristic (ROC). Specifically, the ROC is not as curvilinear as a signal detection theory requires. In addition, the z-ROC is usually curvilinear, as a threshold recollection model requires, not linear, as a signal detection model requires. In Experiment 1, word pairs were strengthened at study, which yielded a curvilinear ROC and a linear z-ROC (in accordance with signal detection theory). This result suggests that associative recognition performance was based on a continuous variable, one that likely consists of either unitized familiarity or continuous recollection. The remember-know procedure and an unexpected cued recall test suggested that the more curvilinear ROC in the strong condition was mainly due to increased recollection. In Experiment 2, word pairs were presented for an old-new recognition decision before being presented for an associative recognition decision. When pairs consisting of items not recognized as having been seen on the list were removed from the analysis, the ROC again became curvilinear, the z-ROC again became linear, and most associative recognition decisions were associated with remember judgments. These findings suggest that the curvilinear z-ROC often observed on associative recognition tests results from noise, as a mixture signal detection model assumes, and that recollection is a continuous process that yields a curvilinear ROC that is well characterized by signal detection theory. (Contains 5 tables and 9 figures.)