Using the technique of free association in an experimental setting, this author and colleagues previously found differences in the quality of memory traces involved in the production of REM and sleep onset (SO) dreams. The present study aims to answer two questions raised by those results: (a) Must the associative session be temporally contiguous to the dream production to gain access to the sources of the dream? (b) Are the differences between SO and REM associations due only to differences in structure and content of the two types of dream reports? Free associations with the same dream collected immediately after an experimental awakening were compared with those recalled 2 months later. A first group of results supports the hypothesis that free associations are sensitive to the proximity to the amount of dream production. A second group shows that the differences between SO and REM dreams noted in the previous research are not attributable to differences in formal characteristics of the report.