Abstract:How to predict rock burst according to electromagnetic emission observation data has been always the hot topic in this research field. The methods more often used now are critical value method,synthetically index method and index change error method,etc.. But all these methods lie stress only on the superficial change of data and overlook a lot of features of rock burst and useful information which are concealed and hidden in observation time series. Pattern recognition extracts the feature value of time domain,frequency domain and wavelet domain in observation time series to form multi-feature vectors,use Euclidean distance measure as separable criterion between the same type and different types to compress and transform feature vectors,applies Fisher criterion to form pattern recognizer for dangeorous recognition. The pattern recognizer uses feature vectors being compressed to carry out training and study and gets the structure parameter:discriminate coefficient(weight value) and discriminate threshold. It has become the pattern recognition system with stable function. It is proved by prediction of test sample that predicting precision is prior to traditional predicting methods such as critical value method and so on.