The global effect of road traffic accident has led to numerous efforts by stakeholders to improve on road safety records at different levels. Countries set fatality reduction targets to be achieved over a certain period of time. An efficient scheme for fatality reduction comes with the need to have a fair idea of what the figures will be in the future. Fatality prediction models are used to predict the likely number of fatality over a specified time period. For the Malaysian case, several models are available including Negative Binomial Regression Models, Smeed's Law Models, and ARIMA model. In this article, Generalized Estimating Equation (GEE) is used to estimate road fatality using selected exposure variables. Population, Road Length, Vehicles involved in crashes and Mobile cell subscription per 100 people are found to be significant in predicting annual road fatality. GEE with negative binomial distribution was found to be suitable for use in short-term road fatality prediction modeling. A new exposure variable is proposed, tested and found satisfactory. The model developed was found to be reasonable when compared against similar existing models. © IAEME Publication.