This paper provides a statistical assessment of the effect of occupancy on the rollover propensity of passenger vehicles such as cars, SUVs, minivans, pickup trucks and 15-passenger vans. A logistic regression model has been built to predict the probability of rollover as an outcome of a single vehicle crash, based on occupancy as well as various other vehicle, crash and driver-related factors. The model uses all police-reported crash data from selected states over the period from 1994 to 2001 from NHTSA’s State Data System (SDS). The metric used to compare the relative risk of rollover among the vehicles is the probability of rollover conditional on a single vehicle crash having occurred. A binary logit model is estimated using the Maximum Likelihood (ML) approach. The resulting parameter estimates and test-statistics are used to assess significance of the explanatory variables and to estimate the probability of rollover for plausible scenarios. The analysis has shown that occupancy, along with speed and road geometry, has significant effect on rollover propensity. While the overall pattern points to an increasing risk of rollover with increasing occupancy in all passenger vehicle categories, the magnitude of increase varies significantly among the vehicle classes. In fact, the increase in the modeled risk of rollover from nominal (driver only) occupancy to full occupancy is most pronounced for 15-passenger vans followed by Minivans, SUVs, Pickup Trucks and Cars. Apart from the relative risks at nominal and full payloads, there is also a wide disparity in the predicted probabilities of rollover at various occupancies between the vehicles. In fact, on high-speed roads at full occupancy, 15-passenger vans depict the highest risk of rollover, followed by SUVs, Pickup Trucks, Minivans and Passenger Cars, in that order. Charts depicting predicted probabilities by occupancy for various hypothetical scenarios of crash factors are presented for each vehicle class.