This is a critical review of the analytical models developed by the U.S. Army to predict training device effectiveness. The Simulation Systems Technical Area of the U.S. Army Research Institute for the Behavioral and Social Sciences performed the research. Special consideration is given to a family of models known collectively as TRAINVICE. These four models differ in several important ways, although they are all intended to provide an index of transfer potential. The variables considered in the calculation of the indices are assigned different degrees of emphasis in each model. In selected cases a variable is completely ignored: e.g., the instructional features of a training device. The preparatory analyses employed to estimate the values of each variable change from model to model. The combinatorial procedures used to calculate an index of effectiveness from the variable values are different in each model. The principal points of difference among the various predictive models are presented to facilitate a comparative assessment of their merit. Results of the analysis and subsequent validation will provide a basis for the selection and refinement of one or more of the models for incorporation into a defined set of procedures for specifying training device effectiveness. (Author/DWH)