This article presents a four-component instructional design model for the training of complex cognitive skills. In the analysis phase, the skill is decomposed into a set of recurrent skills that remain consistent over problem situations and a set of nonrecurrent skills that require variable performance over situations. In the design phase, two components relate to the design of practice; they pertain to the conditions under which practice leads either to rule automation during the performance of recurrent skills or to schema acquisition during the performance of nonrecurrent skills. The other two components relate to the design of information presentation; they pertain to the presentation of information that supports the performance of either recurrent or nonrecurrent skills. The basic prediction of the model is that its application leads to “reflective expertise” and increased performance on transfer tasks. Applications of the model that support this prediction are briefly discussed for the training of fault management in process industry, computer programming, and statistical analysis.