We evaluated STREAK and the univariate signal detection model of Remember/Know (RK) judgments in terms of their ability to fit empirical data and produce psychologically meaningful parameter estimates. Participants studied pairs of words and completed item recognition tests with RK judgments as well as associative recognition tests. Fits to the RK data showed that the univariate model provided a better fit than STREAK for the majority of participants. Although associative recognition relies primarily on specific memory evidence for the association, both the global and specific memory strength estimates from STREAK strongly correlated with associative recognition performance, and the correlation was actually nominally stronger for global strength. Thus, STREAK did not fit the data as well as the univariate model and did not produce interpretable estimates of global and specific memory strength. The success of the univariate model suggests that RK judgments are based on a single conglomerate of all available memory evidence and do not reflect qualitatively different memory systems or processes.