Empirical evidence from cognitive learning research suggests guidelines for structuring instructional information in a manner which is consistent with the way people process it. A cognitive engineering process based on these guidelines was used to restructure the content of an existing embedded training lesson. A heuristic was developed to select and adaptively sequence the most appropriate exercises for the individual trainee based upon the individual's history of performance. An experimental evaluation of the effectiveness of the cognitively engineered instructional content and the adaptive sequencing strategy was conducted. Trainees using the cognitively engineered lesson made 65 percent fewer errors on a performance test than the control group. The results also demonstrate that adaptive exercise sequencing will increase both the effectiveness and efficiency of computer based training. These results are discussed in the context of the application of intelligent tutoring system research to embedded training.