The risk of injury for particular car makes and models was examined in relation to their corresponding crash performance in the 35mph New Car Assessment Program's (NCAP) frontal barrier tests. Texas State accident data for the years 1980 to 1982 were used to determine the risk of driver injury in single-vehicle, fixed-object car collisions. The risk of fatal or incapacitating injury was modeled using logistic regression techniques so the effect of confounding factors such as car mass, age of driver. crash severity, and restraint use could be effectively controlled. The NCAP test results were entered in the model by using the values of Head Injury Criterion (HIC), chest deceleration (CD), and femur load as independent variables. Separate results are presented for restrained and unrestrained drivers. After controlling for the confounding effects, for restrained drivers the risk of driver injury was reduced by between 80 and 87 percent for cars with the best NCAP test results compared to cars with the worst results. The corresponding reduction for unrestrained drivers was between 21 and 36 percent. To examine the possible biomechanical causes of these statistical results, the 16mm films from the NCAP crash tests were analyzed to determine whether there were discernible differences in the trajectories of drivers for cars with good NCAP test results compared to cars with poor results. This analysis showed an obvious pattern of increasing interaction with the steering assembly with increasing HIC and CD values.