In traditional video compression standards, the quantizer plays an important role. It is the same for the distributed video coding (DVC), which can be thought to consist of a quantizer followed by a Slepian-Wolf encoder. The quantizer has dominant effect on the quality of the reconstructed video. In this context, this paper proposes quantization scheme for DVC based on a modified Lloyd-Max algorithm. The reconstruction values and partitions of quantizer are designed online for each Wyner-Ziv frame. The training set is extracted from the previous frame and a corresponding correlation model is computed respectively in the encoder and decoder. It is shown that at the cost of a slight higher error rate of the turbo code, the simple nonuniform quantization scheme will lower the quantization error in the reconstruction step, and on the whole improve the rate distortion performance.