Model initialization in soil organic carbon (SOC) turnover models has often been described as a crucial step in making future projections. Model initialization by the spin-up of pools of SOC (model equilibrium run) has been questioned, because equilibrium has to be assumed. Measured SOC pools are independent of model assumptions and are thought to reflect better real site conditions. It has been suggested that model initialization with measured SOC fractions could provide an advantage over model spin-up of SOC pools. In this study we tested this suggestion in relatively undisturbed native grasslands in Australia. We tested the Rothamsted SOC turnover model (RothC) under climate change at 12 sites with three different initialization methods, viz. model initialization with (i) spin-up of model pools with inert organic matter (IOM) pool size calculated from a regression equation, (ii) spin-up of model pools with measured IOM and (iii) all pools estimated from measured fractions. Averaged over the sites and initialization methods, maximum absolute variations (absolute differences in projected SOC stocks expressed as a percentage of initial 2008 SOC stocks) as well as averaged absolute variations throughout the projection period were very small (2.2 and 1.6%, respectively). Averaged across the sites, there were no significant differences in projected grassland SOC stocks under climate change after 93 years of simulation with model initialization by different methods and averaged absolute variation was only 1.6% across initialization methods. These findings suggest that in a relatively undisturbed land-use system such as native grassland, projections of SOC under climate change are relatively insensitive to the model initialization method.