Bicycle is an alternative transportation mode promoted for health benefits. However, cycling is somewhat risky. In France we estimate an annual incidence rate of 70 injured per 100,000 inhabitants. Most of the existing cycling safety studies are police data based which is inappropriate for cyclist safety research because it is biased on accident type (single vs. multi vehicle crash) and registers up to 10 times less cyclists than a road trauma registry on the same area. That is why in this work we identified injured cyclists in 2009-2011 through a road trauma registry, and conducted a postal survey on them. The survey study sample contains 1078 injured cyclists. This article exposes a Typology of Bicycle Crashes (TBC) construction. In this typology, a multifactorial approach is adopted and the Partitioning Around Medoids (PAM) algorithm is used. Seventeen crash configurations compose our TBC: 7 utilitarian trips, 3 leisure rides and 7 sport practices. TBC has been constructed with 35 variables; the most discriminatory variables are "collides with another road user", "commutes", "practices sport" and "aged 60 and over". Each accident configuration is described by the proportion of its main characteristics in comparison with the one in the overall survey sample. Some factors such as "avoids another road user" (10%), "bad weather" (13%), "riding at night" (14%), "off-road ride" (20%), "slips on the road" (40%) are other accident characteristics discriminant for collisions with another road user or with an obstacle. Injury severity in the 17 configurations is explored.