Both generative learning tasks and retrieval practice tasks can serve as a beneficial follow-up to an initial study phase in which learners have studied new learning material. However, research that compares the effects of these 2 types of learning tasks is scarce. Therefore, it is widely unknown whether, and if so under which conditions, the one or the other type of task is better suited to optimizing learning outcomes. We hypothesized that in learning from text, the effects between generative learning and retrieval practice tasks depend on the cohesion and elaboration of the text. To test this prediction, in two 2 × 2-factorial experiments we varied whether university students were prompted to engage in retrieval practice (with vs. without) and generative learning activities (with vs. without) after an initial study phase in which learners read an expository text that was either of high cohesion and elaboration (Experiment 1) or of low cohesion and elaboration (Experiment 2). When the expository text was of high cohesion and elaboration, engaging learners in retrieval practice was beneficial, whereas engaging learners in generative activities was not. By contrast, when the learning material was of low cohesion and elaboration, only engaging learners in generative learning activities was beneficial. Furthermore, in these circumstances engaging learners in generative activities lost its effectiveness when learners were also engaged in retrieval practice. We conclude that generative learning tasks and retrieval practice tasks serve different functions and, thus, differ in the state of learners' mental representations in which they are most beneficial.