Drought is one of the most serious natural disasters and a threat of food security. It is extremely important to have accurate and timely early warning of drought for prevention and reduction of agricultural disasters. Current method of drought early warning was characterized by weak mechanism, low spatial resolution and lack of crop growth information. A daily crop growth and soil moisture dynamics simulation system using a soil water dynamic model in northern winter wheat zone was constructed. Based on this soil moisture simulation system, a drought early warning system for winter wheat was constructed. Daily drought early warning products from April to May in 2018 were generated by using this warning system. The results showed that forecast accuracy of soil relative moisture in next 10 days was high since its coefficient of determination was 0.63~0.91 and its root mean squared error was 5.6%~18.2%. The closer soil moisture forecast was in time, the more accurate was. Forecast accuracy of drought grades for severe and extreme drought was higher than those of mild and moderate grade. This drought warning system satisfied the needs of drought early warning in national agro-meteorological service department in China. The research result would provide scientific basis for prevention and reduction of agricultural disasters and national drought early warning services.