A novel diagnosis approach for proton exchange membrane fuel cell (PEMFC) systems is proposed in this paper. Different fault conditions can be classified based on the patterns of stack voltage fluctuation, which can be extracted by the autoregressive model (AR model). The proposed method focuses on the stack voltage fluctuation over time, thus it is more practical and less complex as only the stack voltage needs to be collected. The AR model is employed to extract voltage fluctuation features, and then several widely applied classifiers are applied to classify fault conditions. Experiments are carried out to demonstrate the effectiveness. Those faults are introduced by the adjustment of anode stoichiometry, cathode stoichiometry, relative humidity level, and the cooling circuit temperature. The diagnostic accuracy for single-fault conditions is 99%, while it is 93% for multi-fault conditions. Also, compared to the singularity analysis method in our former research, the proposed method is more time-saving. Moreover, the voltage sampling frequency and sample window length are adjusted to research the diagnosis effectiveness, which is studied and discussed for the first time.