In order to gather as much information as possible on road crashes and outcomes, routinely collected police reports of traffic accidents and hospital discharge files were individually matched or "linked" using a computerised iterative procedure on name-identified data from both sources. The two groups of linked and unlinked hospital records were compared. Within the linked dataset, a comparison of like variables was made and showed good agreement between the two sources on accident type and road user type. However, police-reported levels of injury severity were shown to be less reliable. In addition, the proportion of hospital inpatient records that linked to a police record was found to be influenced by several factors. The overall linkage rate from hospital to police was 64% but varied from 29% for motorcyclists in single-vehicle accidents to 79% for motor vehicle drivers. The linkage rate increased with increasing levels of injury severity and was substantially lower for casualties of certain ethnic groups. It was deduced that for most instances where a hospital record did not link to a police record, the crash had not been reported. These findings confirm that there was considerable underreporting of hospitalised road casualties to the police and that the extent of underreporting was greater for those less severely injured.