The present study aims at fostering undergraduate medical students' clinical reasoning by learning from errors. By fostering the acquisition of "negative knowledge" about typical cognitive errors in the medical reasoning process, we support learners in avoiding future erroneous decisions and actions in similar situations. Since learning from errors is based on self-explanation activities, we provided additional prompting procedures to foster the effectiveness of the error-based instructional approach. The extent of instructional support in a web-based learning environment with erroneous clinical case examples was varied in a one-factorial design with three groups by either presenting the cases as (a) unsupported worked examples or by providing the participants with (b) closed prompts in the form of multiple-choice tasks or (c) with open reflection prompts during the learning process. Despite significant learning progress in all conditions, neither prompting procedure improved the learning outcomes beyond the level of the unsupported worked example condition. In contrast to our hypotheses, the unsupported worked example condition was the most effective with respect to fostering clinical reasoning performance. The effects of the learning conditions on clinical reasoning performance was "mediated" by cognitive load, and "moderated" by the students' self-efficacy. Both prompting procedures increased extraneous cognitive load. For learners with low self-efficacy, the prompting procedures interfered with effective learning from errors. Although our error-based instructional approach substantially improved clinical reasoning, additional instructional measures intended to support error-based learning processes may overtax learners in an early phase of clinical expertise development and should therefore only be used in moderation.