The classified side-match vector quantizer, CSMVQ, has already been presented for low-bit-rate image en-coding. It exploits a block classifier to decide which class the input vector belongs to using the variances of the upper and left codewords. However, this block classifier doesn't take the variance of the cnrrcnt input vector itsclf into accounl, This let-ter presents a new CSMVQ in which a two-level block classifier is used to classify input vectors and two diffrent master code-books are used for generating the state codebook according to the variance of the input vcctor. Experimental results prove the cffectivcness of the proposed CSMVQ.