The stability of forecast refers to the consistency of multiple timeliness forecast conclusions issued at different times in the same period. It is an important aspect of the model system, and the large instability will cause trouble to users. In order to understand the stability of common operational models, the relative standard deviation index is used to calculate the magnitude of precipitation fluctuation, and the Flip-Flop index (FFnorm) is improved to measure the reversal degree of precipitation forecast trend in this paper. Besides, the stability of precipitation forecast in two global models (ECMWF, NCEP-GFS) and three regional models (CMA-MESO, CMA-SH9, HHUPS-ST) in six climate zones in China， is discussed on the basis of the two cases of the observed precipitation and the rainstorm and the above. The results are as follows. When there is precipitation, the relative standard deviation of multiple timeliness precipitation forecasts of global models is smaller than that of regional models. That is, the fluctuation of model precipitation forecast is small. The volatility of each model for the western part of the southwest region, the eastern part of the northeast region and southern part of South China is smaller, while that for the western part of the northwest region is larger. In terms of the change trend of multiple timeliness precipitation forecast, the stability of CMA-MESO, NCEP-GFS and ECMWF is better in the two cases, and the index of FFnorm is smaller than that of HHUPS-ST and CMA-SH9. Among them CMA-MESO has a more prom-inent stability effect in the forecast for Southwest China and parts of South China. The index of CMA-SH9 is the largest and the model forecast is unstable. Each model has a relatively large index in the middle and lower reaches of the Yangtze River, and the stability of multiple timeliness forecast is poor. When there is precipitation, the frequency of stable (monotonically increasing or decreasing or constant) trend of precipitation forecast of CMA-MESO is the most, followed by NCEP-GFS. Under the two precipitation conditions, the precipitation forecast of the two models for each region is that the number of monotonically increasing times is greater than the number of decreasing times, and the monotonically increasing characteristics of the CMA-MESO model are particularly significant. The above characteristics could provide some references for model debugging and forecast decision-making.