Abstract: Early-warning of geo-hazards in China, refers to analyze and forecast the probability of regional geo-hazards induced by meteorological factors(mainly rainfall).It has gradually formed a grading business model since 2003, and is classified into national level, provincial level, city and county level. The technological keys and research difficulties of the business are the study on early-warning model and its reliability problem. Considering time-space scale and technical difficulty, early warning models are modeled mainly on the basis of statistical method. Up to now, two generation models have been formed and include mutually calculation, checking and supply. This paper systematically presents the principle, technical method, business application and improved technique of two generation models. The first generation model is named the implicit statistics models, and also named the critical rainfall criterion. It is established on the basis of different rainfalls in different geological environment areas. The model is widely used for involving only one or a group of parameters. It has been improved in model parameters and correcting criterion recently. But, it has difficulty in showing changes of geo-environment and causes of geo-hazards. The first model is also limited by improvement of warning division, accuracy and renewal of criteria. The second generation model is named explicit statistics models and established by taking into consideration of geo-environment and rainfall multi-parameters. In the model, explicit expression of geo-environmental factors, and the advantages of model principle, spatial precision and improving ability are shown. It significantly improves precision and accuracy of early-warning. The second generation model was developed in 2008 and used in national early-warning business, gradually to provincial business. Both are established on statistical method. The two generation early-warning models were influenced by selection of statistical samples, knowing degree of geological environmental conditions and accurate matching of rainfall data in some degree. Over years, the two generation models are operated successfully and improved continuously. Important contributions to geo-hazards mitigation have been made.