Particles occasionally left in GIS emit signals and the signals have typical patterns according to particle locations in GIS. The aim of our study is to classify the patterns such that a pattern can have a relationship to the proper particle condition due to its location by analysing signal patterns. The signals used here are partial discharge (φ-q-n) data in testing GIS. Using the decision tree method, 1) we can classify various patterns successfully with very low misclassification errors; 2) the classification rule can be easily understood by engineers unlike the neural networks method; 3) the method works well as well as the conventional neural networks type pattern analysis tools even in the case that the dimension of the input vectior is not so large.