Classification of solar radiation zones constitutes the prerequisite for the establishment of regional daily global solar radiation (H) estimation general model. Current zone establishment methods are ordinary based on solar radiation observation stations (SROS) which present a sparse and non-uniform distribution. As a result, the possibility of misclassifications of the stations occurs in cases there are no records of radiation and far away from SROS. Therefore, by using k-means cluster and Support Vector Machine-Genetic Algorithm, a two-step radiation zoning method was proposed in this paper according to: (a) H, sunshine duration, temperature and relative humidity from 98 SROS and (b) sunshine duration, temperature and relative humidity from 562 stations without radiation. The method is capable to combine the SROS and the stations without radiation in the process of classification. Thus, these misclassifications have been effectively reduced and the accuracy of each classification has been significantly improved. Based on the method, five radiation zones have been identified. Concurrently, four sunshine-based models were obtained for each SROS and the analysis of statistical indexes indicated that the cubic models presented the best performances in each zone. According to the best site-specific models and radiation zones, the general models of regional H estimation were developed by introducing the geographical parameters, including latitude and altitude. The comparative results demonstrated that the general models proposed in this paper had better accuracies and can represent the general models for the H estimation of stations without radiation records in China.