Forty-eight drivers answered a set of written questions about their driving style and drove a pre-defined, mixed urban and motorway route under observation. For 20 drivers there was a second observer in the car to check on inter-observer reliability. Relationships were examined between self-reports of driver behaviour and observers' reports, and between both of these and the number of accidents in which the drivers had been involved in the past three years. The results indicated that there was good inter-observer agreement on a number of important variables including speed, calmness, and attentiveness. Inter-observer reliability for other variables was less good; among these were preferred distance to car in front, use of indicators, and aggressiveness. Observers also showed good agreement on overall ratings of driver skill and safety. Observed speed on the motorway correlated well with drivers' self reports of normal driving speed. Observer ratings of calmness correlated significantly with self-reports of calmness. Also observer ratings of attentiveness and carefulness correlated significantly with self-reports of deviant driving behaviour. Observed speed on the motorway showed a clear positive correlation with self-reported accident involvement. The results indicate that self-reports of certain aspects of driver behaviour can be used as surrogates for observational measures, thus providing a convenient extension to the researcher's methodological armoury. One such aspect is speed which appears to play an important role in accident involvement.