Since police selection procedures can be justified ultimately only by reference to the social well being accomplished by selecting good officers, attention must be given to the nature of the criteria, the types of predictors to be used, and the ecological constraints imposed by the behavior settings of the work situation. Training program criteria are easy to obtain, yet these data are not generalizable to street performance, and do not involve the officers' responses to the real job demands. Job survival criteria are to be criticized because if predictive variables are chosen on their ability to predict adaptation to a restrictive environment, then the selection process will perpetuate the abortive adaptation process. Supervisors' ratings of on-the-job performance are limited by the stereotypic assumptions of what attributes and abilities are valuable. Defining the complexity and heterogeneity of police officers' roles implies that good criteria are going to involve relatively complex constructs not well defined by simple ratings of single unit behaviors. Moreover, the twisted-pair phenomenon is such that police selection procedures may be more effective at screening those with a poor probability of successful response to the job demands than at predicting successful adaptation. Selection alone will not determine the behavioral outcome, as complex interactions and social structures will determine the behavior of the new police officer. The prevailing police culture inherently limits the ultimate utility of selection procedures. Efficacious solutions may require challenging previously inviolable assumptions about how the system must work and offering alternatives that may differ radically from the current structure of the police agency. About 35 references are provided.