【Objective】To forecast agrometeorological disasters and their levels as an example of high temperature disaster in Guangdong Province. 【Method】 Due to lack of disaster decision rule and historical disaster level data, high temperature disaster level rules were built using fuzzy clustering algorithm (FCM) based on the meteorological data in the long term. Those rules were concluded from the cluster centers of the key attribute and membership degree matrix according to the maximum membership degree principle. Based on these rules, possible disasters and their levels were predicted by dynamic meteorological data. The back propagation network algorithm (BP) in the absence of disa-ster decision rules was applied to study historical meteorological observation data and synchronous disaster level released by the meteorological bureau. The trained BP network models were accurate to discover the inner rules of disasters, so the BP network models were fit for predicting the possible disasters and their level through dynamic observation of data at many meteorological stations. 【Result and conclusion】 Comparing the results of the two methods of data mining, the neural network is found slightly better than the fuzzy clustering to predict the meteorological disaster level.