Taxi-passenger matching plays a crucial role in modern taxi systems. However, currently, the greedy mechanisms are widely adopted, which may limit the quality of services provided by the systems. In this paper, we first formulate the taxi-passenger matching as a global optimization problem by considering the pickup rate and average waiting time of passengers. Then, we propose a parallel genetic algorithm to solve the problem. New operators, including initialization, crossover, and mutation, are designed specifically for the problem. In addition, we use a divide-and-conquer strategy for dimension reduction. The problem is divided into a number of sub-problems according to the geographical locations of passengers and taxis. Each sub-problem is then solved in a parallel way by a sub-component of our proposed algorithm. Experimental results validate the effectiveness and efficiency of the proposed algorithm. It is able to greatly enhance the quality of services provided by the taxi systems.