Scalable on-demand multicast streaming protocols can effectively deliver video objects. Broadcasting technique is very efficient for popular videos, while multicasting methods achieve better performance if the video is not popular enough. An analytical model is presented to select an appropriate delivery scheme according to the popularity of the video so that the number of server channels used is minimal given the maximum user start-up delay. Numerical results show that not only is the overall bandwidth of the system minimized, but a preset quality of service parameter (the maximum start-up delay) is also satisfied when using the proposed hybrid scheme. And the most important, the proposed algorithm is very easy to implement compared with the existing schemes.