Almost all soil organic carbon turnover models rely on a partitioning of total organic carbon into an inert and a decomposable pool. The quantification of these pools has a large impact on modelling results. In this study several methods to estimate inert carbon in soils, based either on total soil organic matter or physical protection, were assessed with the objectives of (1) minimising errors in carbon and nitrogen dynamics and (2) ensuring usability for sites with marked differences in site conditions. CANDY simulations were carried out by varying solely the method for calculating the size of the inert carbon pool used to initialise the model. Experimental data from Bad Lauchstädt and Müncheberg were used for the simulation. The data were made available for modellers at a workshop held at Müncheberg (Germany) in 2004. The results concerning not only carbon but also nitrogen dynamics were analysed by applying selected statistical methods. It was shown that even in short-term simulations model initialisation procedure may influence the simulation results considerably. Three methods of estimating inert carbon were identified as being the most appropriate. These methods are either based on soil texture or pore-space classes and therefore account for the physical protection of soil organic matter. Thus, physical protection seems to be of major importance. By extending the scope of the investigation into nitrogen dynamics, additional support for the applicability of a selected method was obtained.