Road traffic injuries kill nearly 1.3 million people annually, so enhancement of traffic safety becomes a high priority for both governmental organizations and automobile manufacturers. In addition to safety measures used to prevent accidents, the improvement of occupant protection by adaptive restraint systems may reduce the effects of traffic accidents more effectively than current passive restraint systems. This study investigates numerically the development of a smart restraint system based on pre‐crash classification of occupant posture. A catalog of restraint laws optimized for nine driver postures uniformly distributed in posture space is employed. Performance‐based statistical classifiers are developed to recognize the pre‐crash posture from the signals of a stereo‐vision camera which tracks the driver’s head. In addition to a Bayesian approach used frequently in pattern recognition applications, a new classification approach, called Expected Performance Assessment (EPA), was introduced. The performance of the adaptive restraint system with catalog controller (RSC) was investigated using crash simulations with driver on different 200 pre‐crash postures. The highest level of injury reduction (28.2 %) compared to the nominal restraint system (RSN) optimized for the nominal posture was obtained using a k‐NN classifier and catalog with 8 restraint laws. Improved performance is expected in future studies by expanding the number of restraint laws, experimenting with different restraint parameters, and exploring different sensor signals (features) which may improve classifier accuracy and effectiveness