Internet applications are developed rapidly. It is increasingly important for router itself to improve the ability to deal with networking congestion. Traditional Poisson model is unfit for Internet networks with burst flow. But self-similarity model suitable for Internet networks has not been used widely in practice because of its complex model and complicated calculation. By describing the practical buffer performance in routers, a new fuzzy congestion control model based on queues and a congestion control algorithm based on the model are presented. In the algorithm, all kinds of packets are firstly classified into queues according to their own priorities. Then the buffer state is divided into three phases, including normal, congestion avoidance, and congestion according to their buffer usage ratio. The three phases are crossover each other because of their fuzziness. Then by combining the whole congestion control, with the part congestion control, the fuzzy algorithm is carried out. Theoretical analysis and NS stimulation results show that the proposed algorithm has better networking performance in the fairness of all connections, compared with the traditional schemes, especially keeping from being affected by the connections with congestion. It really improves the routers’ ability to deal with network congestion.