The relation ofthe recognition failure ofrecallable words to overall recognition rates is largely invariant across conditions that influence both recall and recognition separately. In two experiments, the influence of the integration of the members of A-B word pairs on this relation was investigated. In Experiment 1, it was found that deviations of observed recognition failure from predictions ofthe Tulving-Wiseman function (Tulving & Wiseman, 1975) were produced by shallow, nonsemantic encoding. In Experiment 2, the association of category-to-instance pairs was varied. It was found that weak associates caused larger deviations of observed recognition failure from predicted reeognition failure than did strong associates. Such results suggest that a strongly eneoded assoeiation between eue and target elements of A-B pairs is a necessary condition for the adherenee of data to the Tulving-Wiseman funetion. The implieations of these findings for general models of memory are discussed.Recognition failure of recallable words, or, simply, recognition failure, is found in experiments in whieh subjects study A-B word pairs and eomplete sueeessive reeognition and eued reeall tests. Reeognition failure is the phenomenon that aproportion of words reealled on the cued reeall test are not reeognizable by the same subject, and it is typically measured by the eonditional probability of recognition given recall, P(RnIRc), where Rn stands for reeognition hits and Re for eued recall.' An important finding has been that the magnitude of reeognition failure is predietable on the basis of overall recognition rates by a mathematieal funetion, the Tulving-Wiseman function (Tulving & Wiseman, 1975), which has the following form: P(RnjRc) = P(Rn) +.5[P(Rn)-P(Rn)2]. (I) Recognition failure data adhere closely to this funetion aeross diverse experimental situations, as, for example,