In The Netherlands, the web-based tool called 'Stoffenmanager' was initially developed to assist small- and medium-sized enterprises to prioritize and control risks of handling chemical products in their workplaces. The aim of the present study was to explore the accuracy of the Stoffenmanager exposure algorithm. This was done by comparing its semi-quantitative exposure rankings for specific substances with exposure measurements collected from several occupational settings to derive a quantitative exposure algorithm. Exposure data were collected using two strategies. First, we conducted seven surveys specifically for validation of the Stoffenmanager. Second, existing occupational exposure data sets were collected from various sources. This resulted in 378 and 320 measurements for solid and liquid scenarios, respectively. The Spearman correlation coefficients between Stoffenmanager scores and exposure measurements appeared to be good for handling solids (r(s) = 0.80, N = 378, P < 0.0001) and liquid scenarios (r(s) = 0.83, N = 320, P < 0.0001). However, the correlation for liquid scenarios appeared to be lower when calculated separately for sets of volatile substances with a vapour pressure >10 Pa (r(s) = 0.56, N = 104, P < 0.0001) and non-volatile substances with a vapour pressure < or =10 Pa (r(s) = 0.53, N = 216, P < 0.0001). The mixed-effect regression models with natural log-transformed Stoffenmanager scores as independent parameter explained a substantial part of the total exposure variability (52% for solid scenarios and 76% for liquid scenarios). Notwithstanding the good correlation, the data show substantial variability in exposure measurements given a certain Stoffenmanager score. The overall performance increases our confidence in the use of the Stoffenmanager as a generic tool for risk assessment. The mixed-effect regression models presented in this paper may be used for assessment of so-called reasonable worst case exposures. This evaluation is considered as an ongoing process and when more good quality data become available, the analyses described in this paper will be expanded. Based on these analyses, the algorithm will be refined in the near future.