The interaction between soil organic carbon pools and climate change is an important determinant of future atmospheric CO2 concentrations. Much effort has so far been allocated to manipulative process studies and predictive modelling exercises. Here, we examine the potential for directly detecting predicted changes through repeated soil sampling. Two contrasting benchmark plots were selected in the steppe at the Russian–Mongolian border, where soil organic carbon losses are predicted to be around 10% over the first 50 years of climate change. In both plots, 50 samples were taken to 20 and 30 cm depths. The estimated time intervals before re-sampling by the same method that were likely to prove significant soil organic carbon losses (α=0.05; statistical power=0.90) were 43 and 26 years.