An approach to the study of driver behavior resulting in accidents is described. It involves the integration of statistical and clinical methods for assessing accident causation. Under this approach, clinical data are used to identify and prioritize driver behaviors for study, and to assist in developing operational definitions and data collection protocols. Accident and comparable exposure data on the incidence and circumstances of each unsafe action are collected. These data are then compared and statistically evaluated to determine the frequency of such behaviors, the risk associated with their occurrence, and the magnitude and circumstances of their involvement in causing accidents. The need for such data in the development of countermeasure programs is described, and a preliminary listing of UDAs for study is presented.