A new image coding method using predictive classified vector quantisation (PCVQ) is proposed. The image is first divided into different subsources using a two-stage classification method based on the optimum three-level block truncation coding technique. Each subsource is then vector quantised using different codebooks. Simulation results show that an improvement of about 2 to 6 dB in PSNR performance over the JPEG scheme is obtained at the rate of 0.4375 bit/pixel.