While it is generally agreed that traffic safety on urban arterials is closely associated with operational conditions, analysis of these relationships has been hampered by the absence of continuous measurements of operational variables such as traffic flow. Operational features of both peak and off peak were examined on a total 176 arterial segments from 23 different corridors within specific regions. These operational data, together with road segment characteristics, (e.g., segment length, number of lanes, median type), were used to construct models to estimate crash frequencies under various operational conditions for differing road segments. To account for the spatial correlations among the segments along the same corridors, Poisson-lognormal models with a two-level hierarchy under a Bayesian framework were used. Results showed significant relationships among operational conditions, roadway characteristics, and crash occurrence on these urban arterials. Lower average speeds at the corridor segment level were found to be associated with higher crash frequencies. The implications of using FCD data to assess operational conditions, and the use of hierarchical Bayesian models for predicting crash probabilities under different operational conditions are discussed.