Road safety for fleets of vehicles has been neglected in the insurance literature, mainly because appropriate data and methodology were not available. This article makes a threefold contribution: 1) Produce statistics on current fleets’ road safety offences and accidents using a panel of 20 years of data on truck fleets; 2) relate fleets’ offences to accidents; and 3) identify and classify the riskiest fleets for insurance ratemaking based on past experience in managing road safety. Our main technical innovation to the insurance literature is in the estimation of fleets’ distributions of accidents. For each fleet size (or group of sizes), we estimate the parameters of the negative binomial (NB) distribution of the annual number of accidents according to the characteristics of the fleets, the years, and the number of driver (DRV) and carrier (CAR) road safety violations accumulated in the previous year. When the NB model does not accurately predict the mathematical expectation of the number of accidents of larger fleets, we proceed in two steps. First, we estimate the probability of having zero accidents in a year, and then estimate the negative binomial distribution using the estimated probability of having zero accidents, to weight the zeros of each fleet. To achieve our third objective, we construct risk classes for the vehicle fleets using the predicted accident probabilities obtained from the estimated models. Our results show a substantial heterogeneity between fleets in terms of road safety. This information should be very useful for optimal insurance pricing and better incentives for road safety.