Welding may be detrimental to the male reproductive system. To test this hypothesis, semen quality was examined in 35 stainless steel welders, 46 mild steel welders, and 54 non-welding metal workers and electricians. These figures represent a participation rate of 37.1% in welders and 36.7% in non-welding subjects. The mean exposure to welding fume particulates was 1.3 mg/m3 (SD 0.8) in stainless steel welders using tungsten inert gas, 3.2 mg/m3 (SD 1.0) in low exposed mild steel welders using manual metal arc or metal active gas (n = 31), and 4.7 mg/m3 (SD 2.1) in high exposed mild steel welders (n = 15). The semen quality of each participant was defined in terms of the mean values of the particular semen parameters in three semen samples delivered at monthly intervals in a period with occupational exposure in a steady state. The sperm concentration was not reduced in either mild steel or stainless steel welders. The sperm count per ejaculate, the proportion of normal sperm forms, the degree of sperm motility, and the linear penetration rate of the sperm were significantly decreased and the sperm concentration of follicle stimulating hormone (FSH) was non-significantly increased in mild steel welders. A dose response relation between exposure to welding fumes and these semen parameters (sperm count excepted) was found. Semen quality decreased and FSH concentrations increased with increasing exposure. Significant deteriorations in some semen parameters were also observed in stainless steel welders. An analysis of information from questionnaires obtained from the whole population including subjects who declined to participate indicated an underestimation of effects due to selection bias. Potential confounding was treated by restriction and statistical analysis. The results support the hypothesis that mild steel welding and to a lesser extent stainless steel welding with tungsten inert gas is associated with reduced semen quality at exposure in the range of the Danish process specific threshold limit values of welding.