Presented are the results of an in-service performance evaluation of four guardrail systems: the G1 cable guardrail, the G2 weak-post W-beam guardrail, and the G4(1S) and G4(1W) strong-post W-beam guardrails. The study was focused particularly on estimating the number of unreported collisions and the true distribution of vehicle occupant injuries. The data were collected in portions of Connecticut, Iowa, and North Carolina during a 24-month data collection effort from 1997 to 1999. The collision performance was measured in terms of collision characteristics, occupant injury, and barrier damage. Unreported collisions were counted in all three areas by periodically inspecting guardrails in a specific control section. Because this type of data collection is both time-consuming and sensitive to methodology and human error, its use is not recommended for future in-service evaluation studies. Instead, the use of rates of injury-producing collisions per million vehicle kilometers traveled past the guardrail or other hardware, which can be determined from data on reported injury collisions, hardware inventory, and traffic volumes, is recommended.