Abstract： Aimed at  the smearing effect in contribution plot method and the falt that fault variables cannot be located, this paper proposed a kNN imputation method for fault diagnosis, combining k-nearest neighbor and the contribution plot algorithm. First, PCA was adopted to build an evaluation model and calculate the combined index. Secondly, knearest neighbor imputation method and the control index were combined to extract preliminary faulty variables. Finally, the contribution plots were employed to find the fundamental faulty variables from the preliminary faults. The proposed method can avoid the influence of contribution values of normal variables effectively. A numerical example and Tennessee Eastman (TE) process were given to verify the effectiveness and accuracy of the proposed method, compared with the reconstruction-based method.