Abstract: Using the 2009/2010, 2010/2011 and 2011/2012 Tibet forest fire prevention period (from November to next April) meteorological observations and T639 numerical forecast data, based on artificial neural network BP algorithm (BPM), the forest fire danger rating forest model in 7 days in Tibet was established. The model history matching rate was more than 85%, through the 2012/2013 fire prevention period fire danger rating test report results showed that: the average absolute error in 3 days was less than 0.5, and the average absolute error in 7 days was less than 0.6; comparing with using numerical weather prediction model, the BPM effectively correct prediction of system deviation factor. It indicated that: the BPM had a good prediction results. The model improved on the Tibetan forest fire danger rating prediction to improve forest fire prevention.