The effects of road measures implemented at main road junctions in 1984–1986 were studied on the basis of police-reported accidents that had occurred at the junctions in 1983–1987. The data were obtained from an extensive junction inventory performed by the District Offices of the Finnish National Road Administration in 1988. A total of 325 three-arm and 298 four-arm junctions were included in the study. The number and type of accidents were studied at each junction both before and after the implementation of the measure. The effects of the measure were determined by comparing the observed number of accidents after the measure with the number that would have been expected to occur at the junction if the measure had not been implemented. The expected number of accidents was calculated on the basis of the observed number of accidents in the before period and the expected number of accidents as predicted by an accident model, utilizing at the same time the information on the variation of the number of accidents at similar junctions revealed by the model. Likelihood functions determined for the effects of the measures were used for studying the accuracy of the estimates of the effects. The likelihood functions can also be utilized later when new before-and-after studies are performed. Road lighting, stop signs, signal control, and lowering of the speed limit value were found to decrease the number of accidents. Through-flow junction widenings, additional lanes for turning vehicles, and road widenings, however, did not seem to affect the safety at junctions to any marked extent. If we had not accounted for the regression-to-the-mean effect, these measures would have seemed to have a positive effect on safety. The study method also enabled us to quantify the regression-to-the-mean effect. The magnitude of the regression effect was on average 20%, i.e. the number of accidents would have decreased by 20% at the junctions studied even if the measures had not been implemented. The magnitude of the regression-to-the-mean varied greatly between the different measures.