Abstract: Smelting regenerating is one of the most widely used ways to recycle heavy metal hazardous waste. Under the failure of the flue-gas treatment units of smelting facilities, the exhaust gas emission has the characteristics of significant source intensity, strong randomness of short-term emission diffusion parameters, and strong randomness of the environmental consequences, which poses a great challenge to the accurate monitoring and scientific decision-making in the emergency response. In this regard, a risk assessment method based on Gaussian Plume Model-Stochastic Response Surface Method (GAUSS-SRSM) coupling was proposed to quantitatively evaluate the random distribution and probability characteristics of pollutants under complex source intensities, diffusion parameters and their uncertainties. An enterprise in North China was selected to carry out a case study. The results showed that under the typical climatic conditions of the region, the concentrations of SO₂ and Pb at 0.8-2.2 km and 0.75-1.5 km downwind might exceed the standard. The probability of exceeding the maximum ground concentration limit was 44% and 28%, respectively. The exposure concentrations characterized at 95% confidence level were 0.68 and 0.005 2 mg/m³, which exceeded Ambient Air Quality Standards (GB 3095-2012) by 1.36 times and 1.16 times, respectively. Seasonal differences in wind speed and temperature, etc. led to large differences in pollution and its probability characteristics in different seasons at the same facility. In the case of Pb, for example, the maximum range of potential exceedances in winter differed by 0.6 km from that in summer, with 24% difference in exceedance probability and 0.003 9 mg/m³ difference in exposure concentration. The uncertainty of climate and source intensity made the air pollution consequences have obvious uncertainty, with the maximum uncertainty of 3.85 at 0.5 km downwind. As the distance increased, the uncertainty decreased, with only 1.74 at 3.0 km. As the large degree and high possibility of pollution occurred at 0.8-2.2 km downwind, it was necessary to avoid the deployment of pollution-sensitive equipment or devices in this area, and it should be used as a key area for emergency monitoring after the accident; while the uncertainty of 0.5-1.2 km was large, it was also necessary to overcome the random error by strengthening the monitoring frequency.