Accurate estimation of reference evapotranspiration (ET0) is very important in hydrological cycle research, and it is also essential in agricultural water management and allocation. Using less meteorological parameters to estimate ET0 is necessary in areas with limited data. The ability of random forest (RF) and gene expression programming (GEP) algorithm in modeling ET0 was investigated and compared by using fewer meteorological parameters collected from four weather stations of Duan, Hechi, Baise and Rong’an, in karst region of southwest China, over a fiveyear period (2008—2012). Daily climatic data of the four stations, including maximum temperature (Tmax), minimum temperature (Tmin)，sunshine duration (n), relative humidity (RH) and wind speed (u2) were employed to model ET0 by using FAO 56 Penman-Monteith equation as the reference, and their performances were evaluated using determination coefficient (R2) and root mean square error (RMSE). From the statistical results, the derived RF-based (R2 was ranged from 0.809 to 0.991, and RMSE was ranged from 0.158mm/d to 0.678mm/d) and GEP-based (R2 was in range of 0.830~0.977, and RMSE was in range of 0.225~0.645mm/d) ET0 models were successfully applied to model ET0 with different input combinations. When only the temperature data can be used, the RF models produced satisfactory results (R2=0.875，RMSE=0.546mm/d), which can be used as an alternative to the conventional Hargreaves model. The relative importance of meteorological variables for ET0 can be assessed by RF method, the order of the relative importance of meteorological variables was: Tmax, n, Tmin, Ra, RH and u2. In most cases, the RF models were found to perform better than the GEP models. The results were expected to be useful to guide rehabilitation strategies and agricultural water management in karst region of Southwest China.