This study investigates the road safety of cyclists in urban unsignalized intersections by devel-oping an accident model. The study focuses on finding relationships between different traffic and design variables and multi-party crashes involving at least one bicycle at urban unsignal-ized intersections. The model is set up as a negative binomial distributed regression model with a log-link function. The model includes 149 intersection observations from 144 unsignalized intersections from Aarhus and Odense. Across the 144 intersections included in the model, 257 multi-party accidents involving at least one bicycle were reported. The model is based on a combination of police-recorded accidents and accidents from the emergency departments in Aarhus and Odense. As an exposure to traffic volumes, AADT (annual average daily traffic) has been used. The model includes the AADT for motor vehicles on primary and secondary roads and the AADT for bicycles on primary roads. Traffic volumes have been extracted from Mastra (the Danish Road Directorate’s database) where possible. Where traffic counts were not available from Mastra, traffic volumes were estimated using trip rates or manual traffic counts of 30 minutes. In addition to the three traffic variables, the final model also includes three design variables, which are the presence of separators between the roadway and the bicycle path and secondary traffic islands, as well as the longitudinal gradient of the primary road. The results show that the presence of a separator and secondary traffic islands leads to an increased accident density. Intersections with a longitudinal gradient of more than 20 ‰ in the approach to the primary road are also associated with an increased accident density. The results show a safety-in-numbers effect among bicycles, as the expected number of accidents does not increase proportionally with traffic volume. Thus, a doubling of bicycle traffic volume does not lead to a doubling of accident density.