A study was done to explore the development of zonal- and arterial-level collision prediction models that incorporate characteristics applicable to urban transit planning. A generalized linear modeling approach with a negative binomial regression error structure was employed by using a data set from Toronto, Ontario, Canada. The zonal-level models indicate that vehicle kilometers traveled, bus or streetcar kilometers traveled, arterial road kilometers, bus stop density, percentage of near-sided stops, and average posted speed have significant associations with occurrences of transit-involved collisions. The arterial-level models, which were developed for collisions involving all motor vehicles, suggest that average annual daily traffic, transit frequency, segment length, presence of on-street parking, and percentage of near-sided stops are all associated with increased frequency of these collisions, whereas percentage of far-sided stops and average stop spacing are linked with reduced collision frequency. It is evident that models such as those developed in this study can provide transit agencies with decision-support tools for considering safety implications in the strategic and service-planning processes. These models can also be used as a tool to predict future levels of transit-involved collisions for an existing and a new transportation network or arterial route.