Abstract: As a congestion measure, queuing delay has many advantages. But packet loss can not be completely avoided if only queuing delay is used as a congestion measure. When the buffer is not large enough and some packets are dropped, queuing delay can not accurately reflect network congestion. In this paper, a congestion control model is proposed based on queuing delay and packet loss rate, in which dual-mode control is used. When the buffer is large enough, queuing delay is used as a congestion measure, the model achieves stable dynamic properties and weighted proportional fairness among heterogeneous flows. When the buffer is not large enough and packet loss is inevitable, packet loss rate is used as a congestion measure, and the model maintains almost the same performance as when there is no packet loss. The model is stable as it switches between the two modes and the transition is smooth.