Modern motion picture coding heavily relies on intra/inter-frame similari-ties/correlation. Exploitation of possible similarities is done by searching and ref-erencing the most similar previously coded and transmitted frame parts called blocks. In current algorithms, a single reference block is pointed at by a motion vector per a block in the frame currently being coded. In this paper, we evaluate the picture quality advantages of using multiple reference blocks within a previ-ous frame. That is, a single block is represented by a weighted sum of previously transmitted two or three blocks. Weights are tabulated using clustering techniques and represented by a few bits corresponding to weight index. The proposed method leads to a gain of 1.5 dB on average, at the cost of a couple of extra bits to transmit and added complexity for calculating the weights on the encoder side.