Medical errors are an inevitable outcome of the human cognitive system working within the environment and demands of practicing medicine. Training can play a pivotal role in minimizing error, but the prevailing training is not as effective because it directly focuses on error reduction. Based on an understanding of cognitive architecture and how the brain processes information, a new approach is suggested: focusing training on error recovery. This entails specific training in error detection and error mitigation. Such training will not only enable better responses when errors occur, but it is also a more effective way to achieve error reduction. The suggested design for error recovery training is to begin with detecting errors in others. Starting off with highly visible and even exaggerated errors, and advancing to more challenging detections and finally requiring to detect errors within oneself rather than in others. The error mitigation training starts with providing the learners with the correct remedial actions (after they have detected the error). With training, the learners are required to select the appropriate actions within multiple choice alternatives, and eventually are required to generate the appropriate remedial responses themselves. These can be used for instruction as well as for assessment purposes. Time pressure, distractions, competitions and other elements are included so as to make the training more challenging and interactive.