Total concentrations of metals in soil are poor predictors of toxicity. In the last decade, considerable effort has been made to demonstrate how metal toxicity is affected by the abiotic properties of soil. Here this information is collated and shows how these data have been used in the European Union for defining predicted-no-effect concentrations (PNECs) of Cd, Cu, Co, Ni, Pb, and Zn in soil. Bioavailability models have been calibrated using data from more than 500 new chronic toxicity tests in soils amended with soluble metal salts, in experimentally aged soils, and in field-contaminated soils. In general, soil pH was a good predictor of metal solubility but a poor predictor of metal toxicity across soils. Toxicity thresholds based on the free metal ion activity were generally more variable than those expressed on total soil metal, which can be explained, but not predicted, using the concept of the biotic ligand model. The toxicity thresholds based on total soil metal concentrations rise almost proportionally to the effective cation exchange capacity of soil. Total soil metal concentrations yielding 10% inhibition in freshly amended soils were up to 100-fold smaller (median 3.4-fold, n = 110 comparative tests) than those in corresponding aged soils or field-contaminated soils. The change in isotopically exchangeable metal in soil proved to be a conservative estimate of the change in toxicity upon aging. The PNEC values for specific soil types were calculated using this information. The corrections for aging and for modifying effects of soil properties in metal-salt-amended soils are shown to be the main factors by which PNEC values rise above the natural background range.