We present a fast predictive search algorithm for vectorquan- tization (VQ) based on a wavelet transform and weighted average Kalman filter (WAKF). With the proposed algorithm, the minimum distortion code word can be found by searching only a portion of the wavelet transformed code book. If the minimum distortion code word found falls within a predicted search area obtained by the WAKF algorithm, the relative address that is shorter than the absolute address for a full search range is sent to the decoder. Simulation results indicate that the proposed algorithm achieves a significant reduction in computations and about a 30% bit-rate reduction, as compared to conventional full search VQs. In addition, the reconstructed quality is equivalent to that of the full search algorithm.