In this paper we examine the endogenous stratification problem in the evaluation of DSM programs and determine the appropriate weighting necessary for consistent estimation. Both the complexity and number of alternative DSM programs create the need for a careful statement of the problems and solutions. We examine two commonly occurring evaluation situations. pure participation models and models of net conservation. A hypothetical numerical example is presented to help elucidate the concepts, and data from a DSM program sponsored by a large U. S. utility is used to examine the effect of weighting. In both hypothetical and actual DSM programs, severe biases resulted when unweighted data was used.