this paper used the niche particle swarm optimization (NPSO) global search ability to optimize the spread parameter of GRNN, which improved the general regression neural network (GRNN) performance; then a urban domestic water demand predictive model was established by basing on NPSO－GRNN. The results showed that NPSO－GRNN fitting and prediction average relative absolute error of Beijing domestic water demand data between 1988 and 2012 were 0.72% and 0.36%, respectively, the fitting and predicted results were lower than the result of BP neural network algorithm. NPSO－GRNN algorithm can be better fitting to the trend of urban domestic water demand in Beijing City, it has higher prediction accuracy and generalization ability.