Accidental pollution incidents have caused a major threat to water safety of drinking water sources. However, few studies have focused on quantitative risk assessment of pollution incidents in a watershed which contains drinking water sources. A coupling model consisting of the Seveso III Directive, SWAT, and MIKE21 models was constructed for risk assessment of sudden pollution incidents at the watershed scale. The potential hazard of risk sources (e.g., industrial enterprises), the vulnerability of risk receptors (e.g., drinking water intakes), and the environmental risk of different sub-regions of the watershed were evaluated by this model. In addition, a case study was applied in Hongfeng Lake watershed (HLW), where the Hongfeng Lake drinking water source is located. The results showed that about 68% of the industrial enterprises in the HLW were potentially hazardous according to the Seveso III Directive, including 5 high hazard enterprises, 13 medium hazard enterprises, and 37 low hazard enterprises, most of which were concentrated in the coal mining, chemical production, and building material industries. The HLW was divided into the Yangchang River watershed (YRW), the Maiweng River watershed (MRW1), the Maxian River watershed (MRW2), the Houliu River watershed (HRW), and the lake area by the hydrological characteristics, among which, the vulnerability index of YRW was the largest. Besides, it was essential to consider the vulnerability assessment of drinking water intakes when conducting an environmental risk assessment in the HLW. Regional environmental risk grade of YRW, MRW1, MRW2, HRW, and the lake area was high, medium, low, low, and none, respectively. The environmental risk assessment results showed good consistency with the pollution characteristics and spatial distribution of industrial enterprises in the HLW. Furthermore, the theory of a three-level prevention system for “risk sources–water body connection–water intakes” was proposed for environmental risk management in the HLW. Overall, the case study in the HLW indicated that the coupling model proposed in this study had a good compatibility for environmental risk assessment of sudden water pollution incidents in a watershed.