The development of a number of error-prone models to select Pell Grant recipients for validation is discussed. The 1983-1984 Pell Grant validation strategy consists of a two-stage approach: selection using Pre-Established Criteria (PEC) followed by selection using Error Prone Modeling (EPM). The database used for model development consists of a sample of 1980-1981 Pell Grant Recipients. The policy question is which students should be selected for various types of validation measures. Eight effectiveness measures are defined, and for each measure an error-prone model is developed that will identify those cases for which the corresponding type of validations will uncover the highest level of error. The data elements include: income, U.S. taxes paid, household size, nontaxable income, liquid assets, spouse income, and dependency status. The eight models are then compared in order to identify the most cost-effective approach to marginal selection for validation. The measures refer only to the payment consequences of discrepancies likely to be uncovered by the corresponding type of validation being used. Detailed appendices include EPM error tables and Automatic Interaction Detector coding categories for predictor variables. (SW)