RUNNING HEAD: Signal Detection and Remember-Know Address correspondence to:

2 In the remember-know paradigm for studying recognition memory, participants distinguish items whose presentation is episodically remembered from those that are known to be old because of their familiarity. A one-dimensional model of this paradigm, based on signal detection theory (SDT), postulates that “remember ” responses are merely high-confidence “old ” judgments. A meta-analysis of 373 experiments, extending those of W. Donaldson (1996) and J. M. Gardiner, C. Ramponi, and A. Richardson-Klavehn (2002), shows that the ROC curves predicted by this model have the wrong slope. We present a new two-dimensional SDT model in which Old items differ from New ones in both global and specific memory strength: the “old ” versus “new ” response is based on a weighted sum of these dimensions according to the theory, and the “remember ” versus “know ” judgment is based on a weighted difference. The sum-difference theory of remembering and knowing (STREAK) makes predictions about several novel kinds of ROC curves. The results of experiments testing these predictions, as well as existin