Performance testing often entails the usage of expensive, time-consuming measures in the quest for determining the level of performance on some desired behavior. It is concluded that a generalizability theory approach to dealing with departures from reality in testing can aid in the establishment of empirically-based choices of measurement strategies. This paper presents a model for assessing the loss of information due to using a measure which may be less realistic, but more feasible, than the desired behavior. The method is based on the concept of generalizability theory. An example is included along with a brief discussion of relevant considerations in performance testing, a background on generalizability theory, and a discussion on decision-making. (Author/DEP)