- Three experiments investigated assumptions of the process-dissociation procedure for separating consciously controlled and automatic influences of memory. Conditions that encouraged direct retrieval revealed process dissociations. Manipulating attention during study or manipulating study time affected recollection but left automatic influences of memory relatively invariant. However, paradoxical dissociations were found when conditions encouraged use of a generate-recognize strategy, violating assumptions underlying the estimation procedure. Use of subjective reports to gain estimates produced parallel results. Easily observed correlations are shown to be not useful for testing assumptions underlying the process-dissociation procedure. A multinomial model produced results that agree with those from the process-dissociation approach.