The associative deficit hypothesis (Naveh-Benjamin, 2000) proposes a pronounced aging-related deficit in associative memory. Semantic knowledge might moderate this deficit. Previous studies on the locus of this associative deficit yielded mixed findings on whether it stems from difficulties in storing or retrieving associations. In this lab study, we used two paradigms with semantically related and weakly related pairs, which enabled multinomial modeling to directly estimate separate storage and retrieval parameters. The pair-clustering paradigm (Batchelder & Riefer, 1980), involved free recall of clusterable words from the same semantic category (e.g., cat and dog) alongside singleton words with no clusterable associate. The free-then-cued-recall paradigm (Rouder & Batchelder, 1998) involved both free and cued recall of weakly associated cue-target pairs (e.g., flame-heat). The behavioral data from our sample of 63 older (60-89 years) and 65 younger (18-32 years) adults showed aging-related memory deficits on all recall tests. The modeling revealed both impaired associative storage and impaired associative retrieval of weakly associated pairs in the free-then-cued-recall paradigm. There was inconclusive evidence in the pair-clustering paradigm with semantic pairs: older adults’ worse recall in this paradigm was primarily driven by a deficit in item memory. There was no indication that cluster storage differed between age groups, but results for cluster retrieval were ambiguous. This suggests that older adults compensate at least during storage processing for declining associative memory by relying more on semantic knowledge.