A method for two time-scales video stream transmission control is proposed, which mainly includes the following three steps. Firstly, a model-based long time-scale bandwidth trend extraction step is proposed to calculate a current bandwidth through a Transmission Control Protocol (TCP) throughput model, and predict a trend of network bandwidth by using an Exponential Smooth Model (ESM) according to the calculated current bandwidth. Secondly, a short time-scale bandwidth fluctuation detection step is proposed to divide a network status into three categories and according to different network statuses, additively increase or multiplicatively decrease the estimated bandwidth. Thirdly, a target bit rate adjustment step based on two asymmetrical thresholds is proposed to set an up-threshold and a down-threshold of bandwidth to avoid frequently adjusting a target bit rate of an encoder. The method can satisfy the requirements of TCP-friendliness, real-time, and smoothness for the video transmission.