Traditional indexing methods face the difficulty of 慶urse of dimensionality?at high dimensionality. Accurate estimate of data distribution and efficient partition of data space are the key problems in high-dimensional indexing schemes. In this paper, a novel indexing method using vector quantization is proposed. It assumes a Gaussian mixture distribution which fits real-world image data reasonably well. After estimating this distribution through EM (expectation-maximization) method, this approach trains the optimized vector quantizers to partition the data space, which will gain from the dependency of dimensions and achieve more accurate vector approximation and less quantization distortion. Experiments on a large real-world dataset show a remarkable reduction of I/O overhead of the vector accesses which dominate the query time in the exact NN (nearest neighbor) searches. They also show an improvement on the indexing performance compared with the existing indexing schemes.