In this paper, we propose a Wyner-Ziv video coding scheme with spatio-temporal side information. In the scheme, temporal side information is generated through motion compensated temporal interpolation, using frames that are already decoded. While spatial side information is created through spatial prediction, using the temporal side information. The spatio-temporal side information are both incorporated into a turbo code based Slepian-Wolf decoder to decode the quantized source, but only temporal side information is used for reconstruction. Extensive simulations over various test video sequences show, compared with a similar codec with only temporal side information, the proposed scheme can achieve more than 20% bit rate savings without sacrifice in PSNR, and hence significantly improves the coding efficiency.