This paper describes the results of an ongoing project in the GM Motorsports Safety Technology Research Program to investigate Indianapolis-type (Indy car) race car crashes using an on-board impact recorder as the primary data collection tool. The paper discusses the development of specifications for the impact-recording device, the selection of the specific recorder and its implementation on a routine basis in Indy car racing. The results from incidents that produced significant data (crashes with peak decelerations above 20 G) during the racing seasons from 1993 through the first half of 1998 are summarized.The focus on Indy car crashes has proven to provide an almost laboratory-like setting due to the similarity of the cars and to the relative simplicity of the crashes (predominantly planar crashes involving single car impacts against well-defined impact surfaces). The recording of the vehicle decelerations in real racing car crashes adds unique, new information on the impact tolerance of the human body. Over 262 recordings have been catalogued to date. Examples of impact recordings are given which are remarkable in terms of the severity of crashes and, in most cases, the resulting lack of significant injuries. A total of 202 cases with peak decelerations above 20 G are summarized. The mean peak rigid body chassis decelerations for the sample were on the order of 53 G. Peak decelerations in excess of 60 G (some as high as 127 G) have been recorded for significant durations in many frontal, side and rear impacts. Associated mean total velocity change was 28.3 mph for the sample, with some velocity changes as high as 60 to 70 mph.Using this data to provide new insights into the tolerance of the human body to impact loading will require associated laboratory testing and modeling to obtain specific information on the loads to body regions such as the head and lower extremities. However, the relatively tight coupling of the driver's torso to the chassis allows direct inferences of the loads on the torso, particularly in side impacts. The data calls into question the use of chest acceleration as an injury assessment criterion in both frontal and side impacts.