Solar radiation is a primary driver for many physical, chemical and biological processes on the earth’s surface, and complete and accurate solar radiation data at a specific region are quite indispensable to the solar energy related researches. This study, with Nanchang station, China, as a case study, aimed to calibrate existing models and develop new models for estimating missing global solar radiation data using commonly measured meteorological data and to propose a strategy for selecting the optimal models under different situations of available meteorological data. Using daily global radiation, sunshine hours, temperature, total precipitation and dew point data covering the years from 1994 to 2005, we calibrated or developed and evaluated seven existing models and two new models. Validation criteria included intercept, slope, coefficient of determination, mean bias error and root mean square error. The best result (R2 = 0.93) was derived from Chen model 2, which uses sunshine hours and temperature as predictors. The Bahel model, which only uses sunshine hours, was almost as good, explaining 92% of the solar radiation variance. Temperature based models (Bristow and Campbell, Allen, Hargreaves and Chen 1 models) provided less accurate results, of which the best one (R2 = 0.69) is the Bristow and Campbell model. The temperature based models were improved by adding other variables (daily mean total precipitation and mean dew point). Two such models could explain 77% (Wu model 1) and 80% (Wu model 2) of the solar radiation variance. We, thus, propose a strategy for selecting an optimal method for calculating missing daily values of global solar radiation: (1) when sunshine hour and temperature data are available, use Chen model 2; (2) when only sunshine hour data are available, use Bahel model; (3) when temperature, total precipitation and dew point data are available but not sunshine hours, use Wu model 2; (4) when only temperature and total precipitation are available, use Wu model 1; and (5) when only temperature data are available, use Bristow and Campbell model.