The signal-to-noise ratio (SNR) available at the receiver significantly affects the performance of Car-to-Car (C2C) communication systems. Thus, accurate models to predict the path loss, in particular for non-line-of-sight (NLOS) scenarios, are essential for the development of C2C systems. Since deterministic models are very time-consuming, also fast-computable stochastic models are of great interest. This paper focuses on the radio propagation for C2C communications at urban crossroads under NLOS conditions. 3D ray-optical path loss predictions are compared with narrow-band measurements at 5.9 GHz. Furthermore, a stochastic propagation model that can be used to estimate maximum communication distances for different data rates is derived.