This paper proposes a segmentation method to analyze waveforms based on tensor analysis. The method directly identifies the non-stationary parts of the waveforms from the instantaneous voltage and current values instead of using the rms sequence values. Unlike the existing methods, the proposed one computes only one segmentation index instead of one for each phase. The segmentation method was tested using both real-world and synthetic waveforms. Real-world data has been provided by the utility ENDESA DISTRIBUCIÓN (Spain) whereas synthetic data have been generated with ATP. Real waveforms correspond to a variety of disturbances leaded by different underlying causes, such as downstream/upstream fault networks, transformer-energization, capacitor-switching and motor starting events.