Arterial roads constitute the majority of the centerline miles of the Florida State Highway System. Severe injury involvements on these roads account for a quarter of the total severe injuries reported statewide. This research focuses on driver injury severity analysis of statewide multilane high speed arterials using crash data for the years 2002 to 2004. The first goal is to test different ways of analyzing crash data (by road entity and crash types) and find the best method of driver injury severity analysis. A second goal is to find driver, vehicle, road and environment related factors that contribute to severe involvements on multilane arterials. Exploratory analysis using one year of crash data (2004) using binary logit regression was used to measure the risk of driver severe injury given that a crash occurs. A preliminary list of significant factors was obtained. A massive data preparation effort was undertaken and a random sample of multivehicle crashes was selected for final analysis. The final injury severity analysis consisted of six road entity models and twenty crash type models. The data preparation and sampling was successful in allowing a robust dataset. The overall model was a good candidate for the analysis of driver injury severity on multilane high speed roads. Driver injury severity resulting from angle and left turn crashes were best modeled by separate non-signalized intersection crash analysis. Injury severity from rear end and fixed object crashes was best modeled by combined analysis of pure segment and non-signalized intersection crashes. The most important contributing factors found in the overall analysis included driver related variables such as age, gender, seat belt use, at-fault driver, physical defects and speeding. Crash and vehicle related contributing factors included driver ejection, collision type (harmful event), contributing cause, type of vehicle and off roadway crash. Multivehicle crashes and interactions with intersection and off road crashes were also significant. The most significant roadway related variables included speed limit, ADT per lane, access class, lane width, roadway curve, sidewalk width, non-high mast lighting density, type of friction course and skid resistance. The overall model had a very good fit but some misspecification symptoms appeared due to major differences in road entities and crash types by land use. Two additional models of crashes for urban and rural areas were successfully developed. The land use models' goodness of fit was substantially better than any other combination by road entity or the overall model. Their coefficients were substantially robust and their values agreed with scientific or empirical principles. Additional research is needed to prove these results for crash type models found most reliable by this investigation. A framework for injury severity analysis and safety improvement guidelines based on the results is presented. Additional integration of road characteristics (especially intersection) data is recommended for future research. Also, the use of statistical methods that account for correlation among crashes and locations are suggested for use in future research.