Strategic behavior plays a key role in fluid intelligence tasks like Raven's matrices. Some participants solve items using the strategy of mentally constructing the answer (constructive matching), which is effective but costly for complex problems; other participants rely on the less accurate strategy of discarding potential answers (response elimination). While this process is relatively well-known, past research hints that intraindividual changes in strategy use may also take place as the task becomes increasingly difficult; however, intraindividual variability in Raven's matrices is poorly understood. The present study aimed to (a) test the hypothesis that participants dynamically shift between strategies during the course of Raven's matrices, as predicted by the literature, and (b) investigate the possibility that these shifts are moderated by individual differences in both ability and motivation. Two samples of 100 participants each completed Raven's advanced progressive matrices, and measures of working memory and need for cognition. The results confirmed that participants tended to turn to the less costly response elimination strategy as the difficulty of the task increased; this variability in strategy use predicted 78% of item-to-item variance in accuracy. Working memory capacity and need for cognition predicted strategy use, and working memory capacity additionally moderated the shift toward response elimination, so that only participants with both high working memory capacity and high need for cognition continued using constructive matching in the later part of the task. (PsycINFO Database Record (c) 2020 APA, all rights reserved).