Intelligent Transportation Systems (ITS) aim at integration of innovative technologies in next-generation vehicles in order to increase the safety and efficiency standards of vehicular traffic fluxes. Although many sensors are currently available in modern cars, mutual data exchange between vehicles in real-time will be crucial in future smart vehicles, but this domain, and in particular bidirectional Vehicle-to-Vehicle (V2V) through Visible Light Communications (VLC), remain relatively unexplored yet. In this paper we present a system for bidirectional data transmission between vehicles through VLC. Specifically, we implement for the first time a bidirectional VLC link using real motorcycle headlights (HL) and rear lights (RL), and we provide a characterization of transmission performances under direct sunlight conditions in an outdoor scenario for realistic distances (up to 30 m) and relative positions between two vehicles. The performance of the VLC system has been evaluated in terms of Signal-to-Noise (SNR) maps and Packet Error Rate (PER), measured in various configurations and for different baudrates (28 kBaud / 57 kBaud). Our result show that a bidirectional error-free communication is possible between two consecutive vehicles for distances up to 12 m, whilst successful transmission could be observed up to 21 m. Finally, we compare our bidirectional intensity and PER data in realistic ( <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">flat</i> ) configuration against the predictions of a physical model which considers non-ideal orientations of the receiver stage through a novel, simple technique that could also be exploited in different VLC implementations. A very good agreement between the experimental data set and the model is found on the whole measurements grid.