Electric discharges seriously threat the safety of high-voltage switchgear. In this paper, a spectrum-based optical method is proposed for hazardous discharge monitoring. A SiPM-based trinal spectral sensor is developed with good performances in terms of sensitivity, defect resolution, and risk evaluation. Experiments carried out on two types of artificial discharges (i.e., partial discharge and arc discharge) demonstrate that the light intensities coupled in the three spectral bands account for different proportions and the ratio among the three components generally experiences a regular change with increase in severity of discharge. The typical spectral ratio values are then acquired for hazard rating of discharge and recognition of discharge types with high confidence.