Rollover crashes, while less frequent than other types of crashes, result in a disproportionately large percentage of the fatal and serious injuries sustained by motor vehicle occupants, including children (Howard, 1995; Esterlitz, 1989). Howard's analysis of U.S. National Automotive Sampling System General Estimates System and the Fatal Analysis Reporting System databases from 1995 through 1999 found that while only 2.2% of children were subjected to rollover crashes, these crashes resulted in 1,832 (28%) of the 6,570 child passenger fatalities. With the transition by U.S. families from traveling in primarily passenger cars to minivans and sport utility vehicles, there is expected to be a significant increase in the exposure of rollover crashes to child occupants due to the higher frequency of rollover crashes with these vehicles (Howard, 2003; Rivara, 2003; Malliaris, 1987). Howard reported that in 1976 only 20% of the light vehicles sold in the U.S. were light trucks and of these 98% were pickup trucks and full-vans. By 1999, light trucks represented 48% of the vehicles sold in the U.S. and 55% of these were sport utility vehicles and minivans. NHTSA reported in CY 2000 the proportion of single-vehicle rollover collisions by vehicle type was 13% for cars, 14% for vans, 24% for pickups, and 32% for sport utility vehicles. The Federal Motor Vehicle Safety Standards do not include any requirement to test child safety seats (CSS) in rollover conditions for motor vehicle use. The authors have investigated hundreds of automobile crashes with child occupants. Review of these investigations and previous research indicates that one type of child safety seat appears to frequently allow ejection during the rollover, the Booster-With-Shield (BWS) child safety seat. To further explore this observation, a series of four dolly rollover tests were conducted comparing the kinematics of a three-year-old child dummy restrained in a BWS to a CSS incorporating a five-point harness. Four passenger cars of the same model were rolled at 48 to 56 kilometers per hour (30 to 35 mph) with a BWS CSS secured in one rear outboard occupant position and a five-point forward-facing CSS in the other rear outboard position. Internal video cameras captured the child kinematics, while external video cameras recorded the overall event. During three of the four tests conducted, the child dummy secured in the BWS was ejected from the CSS. The child dummy secured in the five-point CSS remained fully restrained within the CSS during all tests. The real world crash investigations and dolly rollover testing both indicate that children restrained in BWS CSSs are at risk of ejection during rollover crashes, while children restrained in child safety seats incorporating a well-fitted, five-point harness remain restrained throughout the rollover tests. Conducting dolly rollover tests with child safety seats provides valuable insight into their crash protection capability during rollover crashes.