A running debate in the recognition memory literature pits a long-standing theoretical framework known as signal detection theory against a popular multiple-process theory

A running debate in the recognition memory literature pits a long-standing theoretical framework known as signal detection theory against a popular multiple-process theory based on remember and know responses (Donaldson, 1996; Dunn, 2004; Gardiner & Gregg, 1997; Gardiner, Richardson-Klavehn, & Ramponi, 1998; Hirshman, 1998; Hirshman & Henzler, 1998; Hirshman & Master, 1997; Inoue & Bellezza, 1998; Malmberg, Zeelenberg, & Shiffrin, 2004; Xu & Bellezza, 2001; Yonelinas, 2002). The prototypical version of detection theory involves two equal-variance Gaussian distributions (one representing targets and the other representing lures) and one decision criterion situated along a decision axis. A test item that generates a memory signal that exceeds the decision criterion is declared to be old; otherwise, it is declared to be new (as illustrated in the upper panel of Figure 1). Although the equal-variance detection model is useful for illustrating the qualitative predictions made by signal detection theory, much evidence has accumulated over the years suggesting that a quantitatively more accurate version of the theory is an unequal-variance model in which the standard deviation of the target distribution slightly exceeds that of the lure distribution (as illustrated in the lower panel of Figure 1). An important feature of the standard detection model is that it assumes that recognition decisions are based on a unidimensional strength-of-memory variable. An alternative view holds that recognition memory is based on at least two processes that correspond to distinct subjective states. One process involves recollection of the contextual associations of the original experience, whereas the other process, considered by most to reflect familiarity and by some to reflect semantic memory, does not (Cary & Reder, 2003; Karayianni & Gardiner, 2003; Tulving, 1985; Yonelinas, 2002). Because the subjective states associated with these two memory processes differ, subjects can indicate which process underlies their decision whenever they decide that a test item is old. They typically do this by saying “know” if their decision is based on familiarity (or semantic memory) and by saying “remember” if their decision is based on recollection. The signal detection versus remember/know debate was initiated by Donaldson (1996), who argued that remember and know responses might reflect different degrees of memory strength, instead of qualitatively different memory processes (cf. Knowlton & Squire, 1995). Donaldson suggested that subjects approach the task by adopting two decision criteria, as illustrated in Figure 2. One is situated at a relatively low point on the decision axis (the know criterion), and the other is situated at a relatively high point on the decision axis (the remember criterion). A remember response indicates that the memory strength of the test item exceeds the high remember criterion, whereas a know response indicates that its strength falls above the know criterion but below the remember criterion. If the detection interpretation is correct, then remember/know judgments do not reflect qualitatively different forms of memory, so results based on these judgments cannot be used to effectively investigate dual-process theories of recognition memory. Correspondence concerning this article should be addressed to J. T. Wixted, Department of Psychology, 0109, University of California, San Diego, La Jolla, CA 92093-0109 (e-mail: [email protected]). In defense of the signal detection interpretation of remember/know judgments