Although instruction using expert-generated schemas is associated with higher diagnostic performance, implementation is resource intensive. Learner-generated schemas are an alternative, but may be limited by increases in cognitive load. We compared expert- and learner-generated schemas for learning ECG rhythm interpretation on diagnostic accuracy, cognitive load and knowledge acquisition. Fifty-seven medical students were randomized to two experiments. Experiment 1 (n = 29) compared use of traditional teaching frameworks to expert generated schemas. Participants randomly received either a traditional framework or an expert-generated schema to practice each of two content areas in a crossed design. Learning accuracy and cognitive load were measured during the training phase. Discriminating knowledge and diagnostic accuracy were tested immediately after the training phase and 1-2 weeks after. Using the same methodology, experiment 2 (n = 28) compared use of learner-generated versus expert-generated schemas. In experiment 1, learning from expert-generated schemas was associated with lower cognitive load (13 vs 16, p < 0.001), higher diagnostic accuracy on immediate testing (40 vs 29%, p = 0.018), and higher discriminating knowledge (81 vs 71 %, p < 0.001). Both groups performed similarly on delayed testing (14 vs 8%, p = 0.6). In experiment 2, use of learner-generated schemas reduced diagnostic accuracy during the training phase (55 vs 77%, p < 0.001), with similar performance on the immediate (30 vs 33%, p = 0.89) and delayed (7 vs 5%, p = 0.79) testing phases.. Learner-generated schema generation was associated with increased cognitive load (17.1 vs 13.5, p < 0.001). When compared to traditional frameworks, use of an expert-generated schema improved learning of ECG rhythm interpretation. Participants generating their own schemas perform similarly to those using expert-generated schemas despite reporting higher cognitive load.