The Maryland State Highway Administration (MDSHA) conducts periodical friction surveys on the pavement network to monitor its overall skid resistance condition. It was found from the survey data that pavement friction could be affected by numerous variables: rehabilitation activities, traffic patterns, traffic volumes, survey vehicle test speeds, weather-related factors, and etc. Those variables affect pavement friction condition simultaneously by different mechanisms and in different degrees. It is of importance to investigate the relative significance of those variables in affecting pavement friction so it can be used for friction data interpretation and engineering decision making. This paper applies a multivariate analysis on the friction data using stepwise regression. The stepwise regression provides a measure to assess the relative importance of those variables based on the order of their entry into the model. Cross-validation is then used to examine the effectiveness and generality of the developed model. The engineering implications of the developed model are discussed. The advantages and limitations of using stepwise regression and feasible alternatives for analysis are also discussed.