An effective classification and identification tree for automatically classifying overvoltages based on measured data is built. Firstly， the time domain features are extracted from three-phase overvoltage signals. The set of overvoltage category is classified into two subsets. Secondly， overvoltage signals are decomposed using discrete wavelet transform while other features are extracted from the wavelet transform domain. To make these features more distinctive， overvoltage signals belonging to different subsets each are resampled at different frequencies and decomposed to different resolution levels. Finally， binary classifiers based on the support vector machine are each built at a point on the classification tree and cross-validated using measured overvoltage data. The total identification rate is 95%， indicating that the classification tree can effectively classify overvoltage signals.