Information theory and signal detection theory techniques were used to assess police data reliability by comparing it with MDAI and Level II (technical level) data collected during the Tri-Level Study of Accident Causation. Results indicate the accident level variables reported by the police with least reliability were vertical road character, accident severity, and road surface composition. The most reliably reported data were those concerned with the accident location, date, and number of drivers, passengers, and vehicles. The informativeness of the police reports with respect to driver/vehicle characteristics was practically nil, with the exception of driver age, sex, and vehicle model for which the police were correct most of the time (but not errorless). It was also found that police reports provided very little information regarding the presence of different human conditions and states, vehicle defects and environmental/road deficiencies. The sensitivity of police investigators to all accident causes was low; when causes were categorized into human direct, human indirect (conditions and states) vehicle, and environmental; police were more reliable with respect to human direct causes than to vehicle, environment and lastly, human indirect causes. In the assessment of alcohol presence and involvement a strong and significant difference in reliability existed between male and female drivers, with a lower reliability associated with females. (NTIS abstract)