Abstract: Three daily global solar radiation estimation models (A-P model, ThorntonRunning model and model provided by Liu Kequn et al.) were analyzed and compared using data of 13 weather stations from 1982 to 2012 from three northeastern provinces and eastern Inner Mongolia. After crossvalidation analysis, the result showed that mean absolute error (MAE) for each model was 1.71, 2.83 and 1.68 MJ·m^-2·d^-1 respectively, showing that A-P model and model provided by Liu Kequn et al. which used percentage of sunshine had an advantage over ThorntonRunning model which didn’t use percentage of sunshine. Model provided by Liu Kequn et al. played a good effect on the situation of nonsunshine, and its MAE and bias percentage were 18.5% and 33.8% smaller than those of A-P model, respectively. High precision results could be obtained by using the simple linear model of A-P. A-P model, ThorntonRunning model and model provided by Liu Kequn et al. overvalued daily global solar radiation by 12.2%, 19.2% and 9.9% respectively. MAE for each station varied little with the spatial change of location, and annual MAE decreased with the advance of years. The reason for this might be that the change of observation accuracy caused by the replacement of radiation instrument in 1993. MAEs for rainy days, nonsunshine days and warm seasons of the three models were greater than those for days without rain, sunshine days and cold seasons respectively, showing that different methods should be used for different weather conditions on estimating solar radiation with meteorological elements.