It is important and challenging to make the growing image repositories easy to search and browse. Image clustering is a technique that helps in several ways, including image data preprocessing, the user interface design, and search result representation. Spectral clustering method has been one of the most promising clustering methods in the last few years, because it can cluster data with complex structure, and the (nearly) global optimum is guaranteed. However, the existingspectral clustering algorithms, like normalized cut (NCut), are difficult to use to handle data points out of training set. In this paper, a clustering algorithm named LPC (locality preserving clustering) is proposed, which shares many of the data representation properties of nonlinear spectral method. Yet the LPC provides an explicit mapping function, which is defined everywhere, on both training data points and testing points. Experimental results show that LPC is more accurate than both “direct Kmeans” and “PCA+Kmeans”. It is also shownhat LPC produces comparable results with NCut, yet is more efficient than NCut.