A meta-analysis of proportions of remember, know, and guess responses was carried out on observations from 86 experimental conditions in 23 different recognition memory experiments. Unlike remember and know responses, guess responses revealed no memory for the test items that elicited them. A signal detection analysis of these data showed that A' estimates of the strength of the memory trace depended on response criteria. A' estimates increased significantly when know responses were added to remember responses, and decreased significantly when guess responses were added to remember and know responses. It was guessing, rather than knowing, that was most strongly correlated with overall response criteria. Nor were remembering and knowing correlated significantly. These results do not support a quantitative trace strength model according to which these responses merely reflect different response criteria. They support theories that ascribe remembering and knowing to qualitatively distinct memory systems or processes.