Abstract: Aiming at the features of multiple product processes, nonlinear and non-Gaussian, a fault detection method in batch process based on sparse distance (FD-SD) is proposed. Sparse distance is used to measure the density of training samples around a test sample and to analyze the training samples distribution feature near the test sample. Sparse distance is calculated by cumulative distribution function of sample distance through kernel density estimate function with changed window width. The control limit is calculated by cumulative distribution function of sparse distance, and then the detection model based on sparse distance is built. FD-SD does not needs the hypothesis of variables obeying Gauss and linear distribution and can improve the accuracy and reliability of the fault detection process. From Simulation results in artificial case and semiconductor etch batch process, it is shown that FD-SD can detect fault in nonlinear and multi-mode processes. The validity of FD-SD is proved by the results.