An external insulation of contaminated-insulator assessment method is proposed based on a tri-level data fusion model of combining principal component analysis (PCA) method, artificial neural net (ANN) method and evidence theory in this paper. When contaminated-insulators partial discharge (PD) occur, much effective information obtained from the sound emitted with PD are synthesized to evaluate the external insulation strength of insulators in operation by the studied method. Firstly, nine characteristic parameters that can rapidly reflect the PD process are selected for image-level fusion of PCA to reduce dimension, which gets two new parameters. Then the new parameters are inputted to ANN for feature-level fusion. Finally, the feature-level fusion output is used as the input of decision-level fusion and fused by means of D-S evidence theory for further reducing the uncertainty of assessment. The artificial contamination experiments were explored to verify the proposed method. The result indicates that the proposed model is more precise than the ANN model under the same conditions.