The requirement to achieve and maintain higher and more complex skill levels with fewer dollars faces all modern military systems. One way to accomplish these seemingly incompatible goals is through advanced training concepts, particularly those that move training to the operational platforms. Besides having the obvious advantage of being more accessible than traditional (classroom-based) training, embedded training has the potential to result in better operational performance if developed and implemented properly. The purpose of this paper is to describe a series of research projects aimed at improving the state-of-the-art in embedded training systems. This will be accomplished through a description of a research system exemplifying the latest vision and concepts for embedded training. The system links experiential learning via realistic workstation-embedded problem simulations, with intelligent tutoring via multiple levels of real-time instructional guidance and performance diagnosis. Trainee operator performance is measured against both fixed and adaptive criteria generated from executable cognitive models of expert operators. Cognitive and behavioral diagnosis is dynamically performed based on the differences between trainee behavior and the model-generated criteria. Additional diagnostic inputs from human instructors, via mobile pen-based interfaces, are added to the automated diagnoses. The result is guided practice, in which trainees gain problem-solving practice while receiving real-time and post-problem guidance from both automated and human instructor sources.