Organic carbon levels of 542 soil samples from temperate lowland forest were determined by the original and modified Walkley–Black (WB) dichromate methods and total organic carbon (TOC) analysis. The performance and the lower and upper quantification limits of the WB method were assessed. Variable recovery rates were related to laboratory and field conditions and to the sample composition. The percentage carbon recovered by the original WB method was found to be systematically lower than commonly accepted, and the correction factor was estimated at 1.58 instead of 1.30–1.35. However, a good linear relationship with TOC enabled acceptable prediction of soil organic carbon which was most precise when using the original WB method. Texture class and pedogenetic horizon showed significant differences in recovery. Depending on the modifications of the WB method, recoveries varied significantly between laboratories, explaining up to 29% of the total variation of the topsoil carbon assessment of a site. Low recovery from samples was partly attributed to charcoal and resistant elementary carbon particles. No interference was found to be caused by iron or manganese compounds. In order to use WB carbon data of forest soils, laboratory- and method-specific determination of the recovery rate using a total analyser is strongly recommended. The original WB method was unable to predict reliably forest soil carbon contents higher than 8% TOC.