Several models of memory-based attitude processing were examined in a laboratory experiment. After receiving stimulus information with implications for either one or both of two attitudes (toward a person and toward a behavior with respect to this person), subjects were asked to recall the information and to report their attitudes. Information was received under instructions to form only one of these two attitudes. Consistent with past research, information was subsequently better recalled when it had implications for the attitude being evaluated than when it did not. However, null effects of the instructional set on reported attitudes suggested that subjects relied on their memory-for-attitude (or additional implications) rather than on memory-for-facts. The present findings extend the generality of this conclusion to behavioral attitudes, large stimulus sets, and heterogeneous stimulus items. Low correlations between recalled stimuli and reported attitudes also supported a memory-for-attitude model. However, correlations between behavioral attitudes and recall of behavioral outcomes increased under certain conditions. Results are discussed in relation to recent findings in person perception research.