While SO4(2-) concentrations in runoff are decreasing in many catchments in Europe, present day S output still exceeds the S input for most forested catchments in Europe and North America. Here we report that a large part of the observed SO4(2-) in the runoff at a large-scale catchment study site (the Gårdsjön roof experiment in southwestern Sweden) originates from the organic S pool in the O horizon. Budget estimates comparing soil S pools showed reductions in the S pool of 57 mmol of S m(-2) in the O horizon and 26 mmol of SO4(2-) m(-2) in the mineral Bs horizon after excluding anthropogenic deposition for four years. There was an increase of about 1% per hundred in the delta34S(SO4), value of the mineral soil SO4(2-) between 1990 and 1995 (average and 95% confidence interval of 6.2 +/- 0.6 and 7.7 +/- 0.6% per hundred, respectively), but the delta34S(SO4) values in the E horizon are still much lower than the sprinkler water input of +19.7% per hundred, although the horizon has only a small extractable SO4(2-) pool. After nine years (1991-2000) of artificially supplying S inputs comparable with those amounts supplied by preindustrial rain, the amount of S in runoff still exceeded the input by 30%. This extra 30% corresponds to a loss of 3 mmol of S m(-2) year(-1), compared to the soil S organic O horizon pool of 1098 mmol m(-2) in 1990, suggesting that recovery is delayed for decades, at least.