Abstract: Packet loss led to great degradation of picture quality, when video stream transmitted in error-prone channel. RPS (Reference picture selection) could improve picture quality by selecting suitable reference picture, but RPS greatly increased computation complexity. In this paper one low complex RPS optimal method was proposed. First error concealment steps in decoder was tracked in encoder to reconstruct error picture; second only error reconstructed picture and the preceding picture were taken as optional reference pictures; third picture difference variance was computed by linear prediction when processing R-D optimization. Through above means time of motion estimation was reduced, and then computation complexity was greatly decreased. Experiment result shows the method greatly lower encoders complexity, while only loss small picture quality.