An improved serial computation method is proposed for simulating ceramic grain growth at an atomic scale. The data structure of a ternary tree is used to store orderly the atomic information and speed up the editing of data. Combining with the idea of space partition, an index for space searching is established to reduce the computation time. Simulation results demonstrate that the speed of the serial computation is increased significantly and simulated images are in good agreement with micrographs of practical ceramics. It is illustrated that kinetic exponents and fractal dimensions during the simulation of grain growth are also reasonable in compare with quantitative analyses.