Estimation of the accident severity is a fundamental requirement in accident reconstruction and analysis. Accident severity can be measured in many different ways, but in frontal collisions change of velocity, energy equivalent speed or equivalent barrier speed are frequently used parameters. These parameters are most often estimated from vehicle deformation. It is known, however, that the quality of these estimates is limited if compared with these obtained in laboratory test conditions. To be able to achieve almost the same measurements and measurement accuracy in real-life accidents as in the laboratory, where the acceleration time history is measured, an on-board measurement technique is required. This presentation gives results of tests of a low cost device for measuring the crash pulse for a car involved in an accident, concerning systematic and random error. The device, called the Crash Pulse Recorder (CPR), has been tested previously in several sled tests. The CPR is based on measurment of the movement of the mass in a spring mass system in a collision. A brief description of its construction is also included. The CPR is an integral part of a large accident data collection system including interior and exterior deformation measurements and evaluation of injury outcome. This report presents the results of several full-scale crash tests, undertaken to evaluate the accuracy and precision of the CPR in cars in different impact modes. The tests comprised both offset and angled collisions. Most of the tests were car to car collisions, but barrier tests were also performed. The random error of the CPR was found to be 2.2 km/hr for the ΔV measurements and 0.6 g for mean acceleration.