Traditional objective image quality evaluation measures, such as the MSE or the PSNR, only represent the total difference between the original images and reconstructed images. However in some case, such as there are few large error pixels and many small error pixels in an image, they have not a consistent a result with subjective measure. To cope with this drawback, we propose a new image quality evaluation measure that is called geometry weighted separating block peak signal to noise ratio (GWSB/spl I.bar/PSNR). It corresponds to human visual observations very well. The experimental result shows that this measure is valid, reliable, wieldy and extensible.