To predict failure pattern of masonry wall panel more precisely, this paper improved the discrete series cellular automata (CA) model of masonry wall panels, and put forward two new methods for calculating state values in the CA model. One was the physical CA model based on the finite element analysis of masonry wall panels, the other is the analytical CA model based on the propagating appearance of boundary effect to zones within masonry wall panels. Both new models were verified by examples. The concept of relative variation coefficient was proposed and applied to the discrete series CA model and the analytical CA model. By discussion of the application range of the relative variation parameter, a fundamental technique was presented to predict a serious possible failure pattern of a new masonry wall panel on the base of only one experimental masonry wall panel. Experimental results verified the validity of the method proposed in this paper.