We examined the association between neonatal mortality and ambient sulfur dioxide (SO2) levels in the neighborhood of Mt. Sakurajima, Yamashita public health district of Kagoshima City, during the period between 1978 and 1988. The analysis using Poisson regression models showed that the monthly average level of SO2 was positively associated with the neonatal mortality (P = 0.002). When the SO2 levels were categorized into four groups to estimate the relative risk (RR) of neonatal mortality using the lowest exposure category as a reference, the RR increased with elevated exposure levels (P for trend < 0.001) and was the highest in the group with the highest level of exposure (RR = 2.2, 95% confidence interval; 1.2-4.1). Other than SO2, we also examined the number of eruptions, the amount of ashfall, and the average level of suspended particulate matter. None of these factors was associated with neonatal mortality. Although the present study suggests that increase in SO2 levels has had an adverse effect on neonatal mortality in the neighborhood of Mt. Sakurajima, it is difficult to determine the source of the SO2. Further studies are necessary to elucidate the mechanisms of the excess neonatal mortality probably associated with the volcanic SO2 levels. J Epidemiol, 1999 ; 9 : 344-349