A review of the literature suggests that choice of control group may have affected the policy implications of the major evaluations of governmental training programs. It is argued that the usual evaluation designs underadjust for preprogram differences between trainees and the control group and thus yield biased estimates of program impact. (Author/JKS)