Abstract: The karst area in Guangxi Zhuang Autonomous Region is widespread, with complicated hydrogeological conditions and frequent flooding occurence. The Pingzhihe river basin, which is selected as the study area, originates from Baise City of Guangxi and is a first tributary of the Red River. The main stream of the Pingzhihe river is 81 km long, with a total drainage area of 963 km2. Karst region accounts for about 50% of the drainage area and the average annual precipitation is 1,517 mm. Combined with the structural characteristics of the karst aquifer system and the measured data, in this paper, we analyzed the pattern of groundwater drainage in Pingzhihe river basin and built a rainfall-runoff simulation program suitable for this basin. ArcGIS software is used to construct the digital basin for the study area which is divided into nine sub-basins. At the same time, based on the hydrological data of Tonglao, Liming reservoir and other 8 rainfall stations and Fengwu hydrological stations from 2010 to 2015, the research work has been carried out. Among them, six uni-modal flooding events happened in 2010 to 2015 were selected, from which their recession curves were plotted on semilog coordinates. The curves collectively show that when the logarithm of the flow rate is close to 2.5, the turning point appears. Thus two sections of the curves are linearly fitted and their attenuation coefficients can be obtained from the curve fitting. The results show that the subsurface flow can be divided into rapid and slow underground runoffs. Simplified Boussinesq exponential decay equation was used to calculate the regression coefficients for the two parts of runoffs. The extinction coefficient of rapid underground runoff KQ was 0.988-0.995 and the slow runoff extinction coefficient KL was 0.997-0.999. Based on the traditional Xin’anjiang river model with one linear reservoir, a linear reservoir is added to simulate the rapid underground runoff and the slow underground runoff, respectively, to construct a rainfall-runoff simulation program. The calculation program was compiled with Matlab 2014 software. With the maximum of the deterministic coefficient as the optimization objective, the genetic algorithm is used to determine the model parameters.7 floods from 2011 to 2015 were selected as regular analog floods, and 8 floods were selected as verified floods for runoff simulation. According to the "hydrological forecasting code (GB / T 22482-2008)" assessment criteria, the qualified rate of regular analog was 85.71%, and the average coefficient of determination was 0.846; the qualified rate of verification was 75%, the average deterministic coefficient of 0.893 reached B-grade accuracy. Compared with the traditional Xin’anjiang river model, the average deterministic coefficient and qualified rate of the improved Xin’anjiang river model have been improved. The simulation results show that the simulation scheme is suitable for the Pingzhihe river basin, and can provide reference for the rainfallrunoff simulation and flood forecasting in other similar karst areas in Guangxi.