Heavy metals (HMs) in soil near metal smelting areas have attracted great interest due to their crucial threat to agricultural product safety and human health. We investigated HMs in agricultural soils and crops (wheat and maize) in a Pb/Zn smelting city, central China; analysed the exposure risk of HMs through multiple pathways in a soil-crop system through Monte Carlo simulations; and identified the key driving factors using random forest (RF) model. The results showed that most soil HMs exceeded their background level. Cd had the highest exceeding rate compared with the screening values and the highest exceeding rate (40%, in wheat) compared with the food safety limit. Additionally, Cd had the highest bioaccumulation capacity in both maize and wheat. In identifying the key driving factors of soil HMs, RF had a satisfactory prediction performance for As, Cd, and Pb, with corresponding R2 values of 0.47, 0.51, and 0.60, respectively, with the dominant factors being distance to the Pb/Zn smelter, coal washery, and road area, respectively. However, Cr and Hg exhibited lower R2 values which may be due to their limited influence from the selected anthropogenic, traffic, and industrial variables. Probabilistic health risk assessments through Monte Carlo simulations revealed that the non-carcinogenic risk was around the risk limit (HI = 1), while the carcinogenic risk for both children and adults exceeded the risk limit (TCR = 1E-04), and adults had higher carcinogenic risk than children, which was due to a longer duration of exposure and higher ingestion rate of crop grains in adults. Cd had the highest carcinogenic risk and the highest probability of exceeding the limiting risk. Briefly, levels of Cd in both soil and crops should be controlled, especially for the source of smelting emissions. The results of this study could be treated as valuable references for policymakers to formulate control strategies for soil HM pollution in agricultural areas near Pb/Zn smelting activities.