Instructional systems usually do not limit the time available to a learner for responding to questions or practice items. However, experiments conducted by Robert Tennyson and his research group indicate that with regard to the speed of learning this common practice is less efficient compared with the computer-controlled adaptation of a proper response time limit to the learner's increasing competence during instruction. Until now the theoretical background of these results is not well understood and the effects are only reported by a single research group. In this article two experiments are reported. They are based on recent cognitive theories and are aiming at differences between learner control of the response time and adaptive program control of a response time limit on speed and amount of rule learning. Experiment number 1 (N = 66, 3-group-ANCOVA-design) replicated the results of Tennyson and co-workers: Learning speed is highest under a response time limit which is adapted on-line to the achievement of the student in such a way that there is short time available to respond at low achievement and more time at increasing achievement. Learning speed is slowest under a response time limit which is inversely adapted to increasing achievement. Learner control without any time limit is located in-between. Experiment number 2 (N = 40, 2-group-ANCOVA-design) extends this effect to the overall level or amount of learning within a fixed time period: Students learn more under an adaptive response time limit than under learner control without any response time limit. This effect, however, depends on a successful implementation of the algorithm to adjust the response time limit. Otherwise there is a kind of boomerang effect by which learning is hindered. Furthermore, the results indicate that the effects of an adaptive response time limit are more cognitive than motivational, so that they are in accordance with modern cognitive theories like ACT* and repair theory.