A novel most significant bit (MSB) biased self organizing feature map (SOFM) algorithm, suitable for VLSI implementation, has been proposed for digital still color image compression. By encoding each pixel's RGB values with a negative 128 bias and presenting only its least significant 7-bit magnitude to the SOFM network, 3 to 6 dB improvement in PSNR can be achieved in the reconstructed images. Based on this new encoding scheme, which requires only a simple selective complementation, the entire SOFM network can be significantly scaled down without severely scarifying the visual quality of the displayed images.