“2-Level Weighted Vector Quantization” is proposed as an efficient data compression method for images, without losing edge information. A conventional vector quantizer, such as the LBG-algorithm, does not incorporate a special procedure for retaining edge information which is very important for the under-standing of images. Therefore, the codebook for our quantizer is generated from weighted training vectors according to the contents of the edge information. To suppress degradation of the smooth part of the image, in our procedure the vector quantizer is used sequentially for both the locally averaged and the residual image. To evaluate the quality of the compressed image, we designed an index which is based on the number of edges and the sharpness of each edge in a local area. An experiment was performed to quantize the actual image and the reconstructed image was evaluated according to this index. The results showed that our method achieved a higher score than the LBG-algorithm or the adaptive discrete cosine transform.