Catalyst sintering is a slow complex process which may last for years but is contributed by physical processes of atomic femtosecond scale. In this article, sintering caused silver particle growth is considered to be the result of random migration and coalescence, on the basis of which a random walk model is proposed. Simulation results show that the particle size evolution can be well fitted by generalized power law expression, indicating that the random walk model is valid for sintering processes with different sintering rates and time scales. With the assumption of identical grain sizes, changes in the specific surface intergrain boundary length during particle sintering are predicted by the random walk model and the epoxidation rate is related to the specific surface intergrain boundary length in terms of a selectivity factor. The results elucidate quite well the phenomenon of nonmonotonic change of the selectivity during silver catalyst sintering. (C) 2004 Elsevier Ltd. All rights reserved.