Video streaming is one of the most important applications used in the best-effort Internet. This paper presents a new scheme for multiple source video streaming in which the traditional fine granular scalable coding was rebuilt into a multiple sub-streams based transmission model. A peak signal to noise ratio based stream rate allocation algorithm was then developed based on the transmission model. In tests, the algorithm performance is about 1 dB higher than that of a uniform rate allocation algorithm. Therefore, this scheme can overcome bottlenecks along a single link and smooth jitter to achieve high quality and stable video.