Abstract: In this study, we propose a fault detection method based on the neighborhood preserving embedding-weighted k-nearest neighbors (NPE-wkNN) method to solve complex multimodal process fault detection problems. First, we obtain the manifold structure of data in the feature space using neighborhood-preserving embedding (NPE). Second, we determine the k-nearest neighbor (kNN) set for the K-th nearest neighbor of each sample and calculate the weight of the sample in the feature space. Finally, the weight distances of samples are considered to be the statistics required formonitoring the process quality. The NPE-wkNN method reduces the computational complexity while maintaining the neighbor structure of raw data. Furthermore, the weighted method eliminates the multimodal characteristics of data and improves the fault detection rate of the process when compared with those observed in case of traditional principal component analysis, NPE, kNN, weighted k-nearest neighbor (wkNN) and other methods. The results of the numerical examples and semiconductor etching process can be used to verify the effectiveness of the proposed method.