Multi-surface models are widely used to assess the potential ecotoxicological risk in metal-contaminated soils. Their accuracy in predicting metal speciation in soils with low metal levels was not yet tested. Now highly sensitive analytical techniques are available to experimentally validate such models at low concentration levels. The objective of this study was to test the accuracy of a multi-surface model to predict the Zn2+ concentration and to improve our understanding of Zn bioavailability in low-Zn soils. High-Zn soils were included as controls. Model parameters were determined independently on the basis of earlier peer-reviewed publications. Model output was validated against free Zn2+ concentrations determined with the soil column Donnan membrane technique in a range of soils varying in potentially available Zn, organic matter, clay silicate, and iron (hydr)oxide contents and pH. Deviations between predicted Zn2+ concentrations and experimentally determined values over the whole Zn concentration range were less or equal to the experimental standard error, except for one low-Zn soil. The Zn2+ concentration was mainly controlled by adsorption, where organic matter was predicted to be the dominant soil sorbent. The predicted Zn2+ concentration depends more sensitively upon changes of the reactive Zn pool (application of 0.6, 1.2, 2.4, and 3.6 mg of Zn kg–1 of soil) and organic matter content (±0.2 and 0.4%) than pH changes (±0.5 and 1 pH unit).