Abstract: The most problems of multi-constrained QoS routing are NP-completeness. In this paper, a novel genetic algorithm is given which is applied to QoS routing. The principle of hop-prior is presented to generate an initial population. Simultaneously, two operators of vicinity crossover and forward mutation are brought forward. The primary advantages of this algorithm are shown as below:the population initialization is independent on the metrics in a network, the new personals born of crossover are diversified, and mutation depresses the algorithm devolutions effectively. At last, the simulations on differ-scale networks randomly created demonstrate that this algorithm is effective.