With the increasing number of intelligent networked devices in cities, the control centers of these devices are under increasing communication and computation pressure. Vehicular Ad-hoc Networks (VANETs) routing protocols not only need to address the challenge of frequent changes in network topology but also need to adapt to routing without the assistance of control centers. To this end, this paper proposes a distributed V2V routing algorithm for VANETs based on block Q-learning, which realizes the final routing by connecting the best routing in each block. In each block, the routing relay selection problem is modeled as a Markov decision process and solved using Q-learning. When realizing cross-block routing, we propose an improved Q-value updating formula to integrate the current block routing and neighboring block routing by considering the influence of neighboring block best routing. Simulation results show that the algorithm performs better in terms of average end-to-end delay, hop count, and packet delivery rate than other algorithms.