The setting of the alarm threshold is critical for photovoltaic (PV) fault diagnosis systems. In practice the prior knowledge is generally used to set the alarm threshold with a fixed value. However, the output of a PV array varies with the external environment, and the distribution of the array electrical parameters has obvious volatility and nonlinearity. Therefore, it is difficult to judge the abnormal state of the PV system under all working conditions by setting the alarm threshold to a fixed value. To solve this problem, this paper proposes a new method for setting the alarm threshold of a PV array output. In this paper, the dynamic variation of the PV array output under different environmental conditions is analyzed, quantile method is adopted to set the threshold of the PV array output, and the dynamic threshold is used to realize the abnormal state judgment of the PV array. Finally, the effectiveness of the proposed method is demonstrated by experimental verification in the actual PV system. The proposed method is simple in calculation and can adapt to dynamic changes in the PV output, which provides a new method for the threshold setting of PV system fault diagnosis systems. (C) 2020 Elsevier Ltd. All rights reserved.