This paper analyzes the national records of highway accident experience, for the purpose of evaluating the protection offered by safety belts to car occupants, as a function of seating position, belt type and casualty type. In response to persistent concerns, this investigation emphasizes the need for and the application of controls, necessary to deal with confounding effects and with other limitations inherent in large scale accident data bases. The entire inventory, all eleven years 1975 to 1985, of the Fatal Accident Reporting System (FARS) is the data source for fatal car accidents. Data for all crash severities, whether fatal or not, are supplied by the National Accident Sampling System (NASS), 1979 to 1985. Three methods are used to analyze the data: "Explicit Control for Crash Severity","Subject/Control Occupant Pairs", and "Overall Evaluation of Casualty Reduction". The analysis of the NASS data produces familiar results for safety belt effectiveness, i.e. values scattered around a nominal 50%. A parametric analysis is conducted to determine the sensitivity of safety belt effectiveness estimates to inaccuracies in reporting restraint system status. The FARS data are analyzed by methods that utilize "Subject/Control Occupant Pairs", exposed to similar crash conditions. Results from these analyses indicate the following fatality prevention effectiveness: (a) about 50% for the driver restrained by a Lap & Shoulder belt; (b) about 40% for the front seat outboard passengerre strained also by a Lap & Shoulder belt; (c) a decline of each of the above by about 20 percentage points, when the Lap and Shoulder belt is replaced by a Lap Only belt; and (d) an effectiveness in the range 15% to 25%, with a rather large error, for rear seat passengers, who are restrained exclusively by Lap belts. A similar method is applied for the determination of the fatality prevention effectiveness of ejection avoidance. It is found that this effectiveness rangesb etween 70% and 80%, irrespective of seating position. This is very important given that there are between 6,500 and 8,000 people per year, roughly 2/3 of them car occupants and 1/3 occupants of light trucks and vans, that are killed ejectees. This is the population that may benefit from the high effectiveness of ejection avoidance.