Disturbance classification algorithms are always composed by two sequential processes: the signal feature extraction and the classifier design. A novel disturbance classification algorithm consisting of phase space reconstruction（PSR） and support vector machines（SVM） is presented. At first PSR is applied to construct disturbance signal trajectories converted into binary images by encoding in the next stage. For these binary images， four distinctive indices are proposed to represent discriminative features of different disturbance patterns. Then the obtained features are utilized as inputs into the SVM classifier to realize the automatic classification of power disturbances. Numerical results show that with the merits of less calculation burden， high accuracy and less demand of training samples， the method proposed can effectively classify six disturbance patterns including voltage sags， voltage swells， voltage interruptions， impulsive transients， harmonics and flickers.