FOLLOWING THE DESCRIPTION OF A MODEL TO SIMULATE THE RESULTS OF VARIOUS PLEA BARGAINING STRATEGIES, THE TYPES OF BIAS WHICH CAN BE INTRODUCED INTO THE DATA ARE DISCUSSED. A MAJOR SOURCE OF BIAS COMES WHEN VARIOUS SUBPOPULATIONS OF CRIMINALS ARE WEIGHTED. THIS WEIGHTING CAN PRODUCE PERFORMANCE MEASURES WHICH, WHEN USED AS POLICYMAKING CRITERIA, CAN INTRODUCE UNWARRANTED BIASES AGAINST CLASSES OF OFFENDERS. CAREER CRIMINAL STATISTICS ARE USED TO TEST THE MODEL. BY SIMULATING OFFENDERS INDIVIDUALLY, THE MODEL EXAMINES THE EFFECTS OF HARSH, MODERATE, AND LENIENT PLEA BARGAINING POLICIES ON RECIDIVISM AND CAREET CRIMINAL COSTS. BY ASSUMING THAT A DETERRENT EFFECT DOES NOT EXIST, THE EXPERIMENT WAS ABLE TO SHOW THAT AN AGGREGATE MEASURE OF PERFORMANCE WHICH COMBINES RECIDIVISM AND COST STATISTICS CAN BE MISUSED UNLESS OFFENDERS, REGARDLESS OF THEIR CLASSIFICATION, HAVE EQUAL EFFECTS ON THE MEASURE OF PERFORMANCE. WEIGHTING DESTROYS THESE EQUAL EFFECTS. CRIMINAL JUSTICE SYSTEM POLICY ANALYSTS SHOULD WEIGHT CRIMINAL SUBPOPULATIONS ACCORDING TO THEIR REPRESENTATION IN THE ENTIRE CRIMINAL POPULATION WHEN CONSIDERING POLICY ALTERNATIVES. THIS LARGELY MATHEMATICAL DISCUSSION CONTAINS FORMULAS, TABLES GIVING BOTH INPUT DATA AND RESULTS OF THE SIMULATIONS, AND EXTENSIVE REFERENCES. (GLR)