The particle density of soil (rho(S)) represents one of the basic physical properties of soil. However, measurement of this parameter is not part of common routine soil inventories in most countries. Therefore, pedotransfer functions (PTFs) were developed to calculate rho(S). Here, we used a complex, hierarchically structured PTF to calculate rho(S), separating the soil mineral substance (SMS) into clay, silt and sand fractions as well as separating the soil organic matter (SOM) into heavy-density and low-density fractions. This PTF was recently published, and here, we introduced an additional hierarchical level to consider the particle size fraction-dependent effect of the mineral composition on rho(S). This extended PTF was calibrated and validated using data from soils of 16 German long-term experiments contrasting in soil texture and in soil mineral composition. X-ray diffraction analysis was applied to identify the mineralogical composition of the clay, silt and sand fractions. We fitted the particle densities of the identified minerals occurring in the respective particle size fractions by minimising the squared differences between the measured and predicted rho(S). The model performed very well (RMSE = 0.011 Mg m(-3) soil). According to the mechanistic base of the model and its hierarchical structure, it is very easy to include available information about the composition of any fraction or subfraction of soil mineral substances and to use the model to calculate the rho(S) corresponding to the specific site conditions.