Das and Li proposed a simple method for lossless predictive image coding using two-dimensional space-varying least squares models. It has been observed that the region of support chosen for the predictor need not be as indicated by Das and Li, especially when the choice is motivated by a tradeoff between the performance and numerical complexity of the predictor. A much simpler region of support is suggested and simulation results are provided to show that a better performance is achievable by using the proposed region of support. �� 1995, IEE. All rights reserved.