We compared the effects of item-specific versus relational encoding on recognition memory in the Deese–Roediger–McDermott paradigm. In Experiment 1, we directly compared item-specific and relational encoding instructions, whereas in Experiments 2 and 3 we biased pleasantness and generation tasks, respectively, toward one or the other type of processing. A read condition was tested in each experiment for comparison purposes. Across experiments, item-specific and relational encoding both boosted correct recognition relative to reading, but only item-specific encoding typically reduced false recognition. Signal-detection measures revealed that less information was encoded about critical items after item-specific than after relational encoding. In contrast, item-specific and relational encoding led to equivalent increases in strategic monitoring at test (e.g., use of a distinctiveness heuristic). Thus, monitoring at test was less successful after relational than item-specific encoding because more information had been encoded about critical lures.