We investigate the impact of rainfall on daily road accidents in the metropolitan area of Melbourne, Australia, over 1987-2002. Our analysis from several viewpoints of the accident count, which has been normalised for variation in traffic volume, indicated that the effect of rainfall is multifaceted. Owing to a large non-linear trend a subdivision into three epochs (1987-1991, 1992-1996 and 1997-2002) was made. Nominal daytime and nighttime as well as 3h raw counts were available for the first two epochs only. Generally, the effect of rainfall across the epochs shows a tendency for larger values in autumn with smaller values in spring. For the daily, daytime and nighttime cases there is an approximate 40% decrease in both the volume-normalised dry and wet means from the first to second epoch. Since the second epoch is wetter than the first, and both dry and wet cases are affected in a similar way, then it appears that a non-weather influence is at work. It is suggested that law enforcement measures may be largely responsible. We obtained a conservative estimate of relative risk of an accident in wet conditions based on a matched-pair analysis of 3h dry and wet periods over the first two epochs (1987-1996). As with other studies we find that the risk is greater than unity in almost all cases suggesting that the presence of rainfall consistently represents a driving hazard. Rainfall occurring after a dry spell has an enhanced effect on the volume-normalised accident count as the spell duration increases. The effect of dry spells is more clearly described when broken down by rain class. Generally, there is an increase in the impact of a dry spell when it first rains as the spell duration and rainfall amount increase.