Abstract: The cotton development stages are the important features for the description of cotton individual and population characters. Accurate prediction of cotton development stage is very important for realizing various services on early warning and recovery of meteorological disasters and management strategies for cotton production, but there isn't the prediction function for cotton development stage in nationwide agrometeorological operation. In order to satisfy the service demand, the prediction method of the cotton development stage on the base of the operational data is studied. Cotton is main planted in Xinjiang Uygur Autonomous Region, Yellow River and Yangtze River drainage areas in China, where climate features are different. 50 cotton-produced counties of nationwide agrometeorological observation network are located in above main cotton-produced regions, the observation of their cotton development stages reflect the course and status of Chinese cotton development well. The time for same cotton development stage is different each other in different cotton areas, the difference between years is high by analyzing the variability of cotton development stages. The various factors affecting cotton development for 50 cotton observation stations are very different. So the prediction model of the cotton development stage is built in simple station, simple development period. Among affecting factors of cotton development stages, active temperature accumulation is the most main factor for the whole cotton area, precipitation is also relative in some cotton-produced counties located in Yellow River drainage area. According to above results, operational prediction model based on active accumulated temperature and growth days index for cotton development stages has been built, considering received data available and service characteristics of nationwide agrometeorological operation, combining with the results of weather forecast. Among models, active accumulated temperature and growth days index are refreshed year by year, cotton development stages are refreshed decade by decade, historical or forecast temperature is refreshed day by day, the timely operational dynamic prediction of cotton development stages has been realized in this way. The model results of the historical fitting from 1992 to 2001, extrapolating prediction from 2002 to 2003 and testing prediction in 2004 are as follow comparing with observed cotton development stages. The frequencies within 5 days error are above 80% for the fifth leaves and blooming stages, above 70% for emergence, the third leaves and squaring stages, their average absolute errors are within 4 days for above stages. The frequency within 6 days error is near to 70% and average absolute error is within 6 days for boll opening. The frequency within 10 days error is near to 70% and average absolute error is near to 10 days for stopping growth by integrating two methods of active temperature accumulation and first frost in autumn. These results are satisfied, which could meet the demands of nationwide agrometeorological operation service.