The spatial distribution and temporal process of rainfall are major problems in small mountainous catchments. This study investigated the spatiotemporal characteristics of rainfalls using data obtained from a dense monitoring network (10 gauges) in a small catchment of 48.6 km2. The rainfall process is determined by empirical relations (rainfall-elevation and rainfall-duration relations) with a random fluctuation. The rainfall-elevation relation of event-scale is R = 21.7h + 4.6, the rainfall-area relations are suggested as a Gaussian distribution, and the rainfall-duration relation is a power law function. The process was investigated at 1-min resolution and characterized by index including the onset time, duration, and center time of the rainfall peak period. Analysis revealed that the errors were random and followed a normal distribution. Consequently, a method that incorporates the relations of rainfall amount and the stochastic errors is proposed to simulate the rainfall process. Comparing with the monitored data, both the errors of simulated rainfall amount and peak intensity were in the range of [-30%, 30%]. Additionally, the relations of the simulated and monitored rainfall amount/intensity with the area are both very close. This study marks an attempt to establish a framework for simulation of rainfall at a high temporal resolution (1-min) which is significant to the hydrological hazards forecasting, and realizing the spatial distribution of rainfall within a small catchment.