Optimal binary codes for pixel values are designed for accurate block matching in the fast motion estimation. The conventional binary block matching techniques reduce the computational complexity and the memory bandwidth for the motion estimation, but they degrade the matching accuracy. We find the optimal mapping function between the set of decimal numbers for uniformly quantized pixel values and the set of binary codes, so that the weighted sum of mean squared errors between the absolute differences and the Hamming distances is minimized. Experimental results show that the proposed three-bit binary code set yields about 0.4dB gain over the conventional techniques.