In this article, a new manifold-based process monitoring scheme which incorporates statistical local approach into neighborhood preserving embedding (NPE) is proposed to monitor changes in the local structure of process data. This method not only inherits the ability of NPE to discover the local structure of data but also implements online fault detection by monitoring the local information changes of new observations. Moreover, the incorporation of the statistical local approach allows that no assumptions have to be made on data distribution, since the constructed new monitoring vectors approximately follow multivariate Gaussian distributions. Thus, the confidence limits of two statistics constructed for process monitoring can be easily determined by chi(2) or F distributions. Furthermore, the new developed method can also improve the detection sensitivity significantly. To evaluate the performance of this new monitoring scheme, it was tested in the Tennessee Eastman(TE) benchmark process, and the experimental results have demonstrated its superiority.