In the long-term continuous observation, observed data of global solar radiation (Rs) usually has different degrees of missing. Angstrom—Presscott (A—P) model, which was based on sunshine hours, is a widely used empirical model that has high precision. Three different parameters calibrated methods for A—P model were chosen, which were annually (M1), monthly (M2) and yearly (M3) calibrated by only one time. Radiation data from six radiation stations were used, which were Changdu, Yichang, Kunming, Ganzhou, Hangzhou and Nanning stations. Annual average Rs, annual average reference crop evapotranspiration (ET0) and parameter selection were compared by the three methods, at last, three stations were selected for interpolation accuracy evaluation. The results were as follow: t test results showed that no difference among the three methods in use of data interpolation in Changdu station, severe changes of Rs inter annual was the main reason for poor accuracy. The accuracy of M1 method for data interpolation was higher than those of other methods in Yichang and Nanning stations. When missing data were 40d and 70d, there were no difference between M1 method calibrated and measured values of Rs in calculating ET0, but when the missing data was over 20d, M2 and M3 methods showed significant difference between estimated and measured Rs, its influence on calculating ET0 was the same. Compared with the calibration method for many years, the monthly calibrated method had a little higher precision in estimating Rs, with higher precision in winter and lower precision from June to August in Ganzhou, Hangzhou, Kunming and Nanning stations. Compared with the results before 1990, the parameter a of A—P model was increased dramatically in all stations after 1990, industrial pollution led to an increased aerosol was one of main reasons. In extremely hot year, the accuracy of A—P model was lower than that in normal year, application of the model to irrigation forecast would underestimate crop water consumption, which may bring risk for decisionmaking. The results can be used in A—P model for risk assessment and improving the total radiation time series data of the interpolation precision of missing cases.