Accurate estimation of reference evapotranspiration (ET₀) is of importance for regional water resource management. The present study proposed two artificial intelligence models, random forests (RF) and generalized regression neural networks (GRNN), for daily ET₀ estimation. Meteorological data including maximum/minimum air temperature, solar radiation, relative humidity, and wind speed during 2009 ∼ 2014 from two stations in southwest China were used to train and test the RF and GRNN models by using two input combinations, including complete data and only temperature and extraterrestrial radiation (R_a) data. The k-fold test was adopted to test the performance of models according to temporal and spatial criteria and data set scanning procedures. The results indicated that both local and external RF and GRNN models performed well for estimating daily ET₀, and RF was slightly better than GRNN generally. The high fluctuations in the accuracy ranges justify the importance of applying k-fold test for assessing the model performance, which could avoid drawing partially valid conclusions from model assessments based on simple data set assignment. Overall, both temperature-based RF and GRNN models can accurately estimate daily ET₀, which is helpful for irrigation scheduling in southwest China.