A real-time fault detection and isolation strategy for gas sensor arrays is presented in this paper. To improve the efficiency of the basic kernel principal component analysis (KPCA) algorithm, a sample selection method is utilized to extract the approximate basis of the entire training sample set, and reduce the time consumption on the calculation of the kernel matrix. Further, a novel algorithm for reducing the number of fault direction set candidates is also developed to improve the efficiency of the basic reconstruction-based contribution (RBC) method in fault isolation. The proposed strategy is fully verified in a real experimental system for gas sensor arrays under multiple-fault situations. The results of the experiments illustrate that the proposed strategy provides an efficient scheme for the real-time process monitoring of gas sensor arrays.