This paper presents a Bayesian approach based on a geographic information system (GIS) for areawide identification of hazardous roadway segments for traffic crashes. The conventional observed relative crash risks of roadway segments are based on the observed crash count data and often feature very small or very large outlying relative risks in segments with low traffic volumes or short segment lengths; thus the risk maps based on observed crash risks show a high uncertainty. The Bayesian approach can incorporate both the variability in the observed risks and spatial autocorrelation and use the typical Bayesian borrow of strength from adjacent segments to filter out the uncertainties, smooth the relative risks, and accurately identify and rank roadway segments with potentially high relative risks for crashes. To capture the real safety indications better, this paper also disaggregates different road segment types and utilizes GIS to analyze the crash data and visualize the spatial relative crash risks in three-dimensional views. Houston, Texas, is chosen as the study area, and three crash data sets (all KAB crashes: crashes with fatality [K], incapacitating injury [A], and nonincapacitating injury [B]; KAB crashes under unsafe speeding; and KAB crashes under the influence of alcohol) in year 2000 are analyzed. Results demonstrate that the approach is effective in spatially smoothing the relative crash risks and eliminating the instability of estimates, while maintaining real safety trends, especially for data with a small sample size and large variance. The result provides a sound basis for implementing preventive actions to reduce crashes on high-risk segments.