The terrestrial biosphere is currently thought to be a significant sink for atmospheric carbon (C). However, the future course of this sink under rising [CO2] and temperature is uncertain. Some contrasting possibilities that have been suggested are: that the sink is currently increasing through CO2 fertilization of plant growth but will decline over the next few decades because of CO2 saturation and soil nutrient constraints; that the sink will continue to increase over the next century because rising temperature will stimulate the release of plant-available soil nitrogen (N) through increased soil decomposition; that, alternatively, the sink will not be sustained because the additional soil N released will be immobilized in the soil rather than taken up by plants; or that the sink will soon become negative because loss of soil C through temperature stimulation of soil respiration will override any CO2 or temperature stimulation of plant growth. Soil N immobilization is thus a key process; however, it remains poorly understood. In this paper we use a forest ecosystem model of plant-soil C and N dynamics to gauge the importance of this uncertainty for predictions of the future C sink of forests under rising [CO2] and temperature. We characterize soil N immobilization by the degree of variability of soil N:C ratios assumed in the model. We show that the modeled C sink of a stand of Norway spruce (Picea abies (L.) Karst.) in northern Sweden is highly sensitive to this assumption. Under increasing temperature, the model predicts a strong C sink when soil N:C is inflexible, but a greatly reduced C sink when soil N:C is allowed to vary. In complete contrast, increasing atmospheric [CO2] leads to a much stronger C sink when soil N:C is variable. When both temperature and [CO2] increase, the C sink strength is relatively insensitive to variability in soil N:C; significantly, however, with inflexible soil N:C the C sink is primarily a temperature response whereas with variable soil N:C, it is a combined temperature-CO2 response. Simulations with gradual increases of temperature and [CO2] indicate a sustained C sink over the next 100 years, in contrast to recent claims that the C sink will decline over the next few decades. Nevertheless, in using a relatively simple model, our primary aim is not to make precise predictions of the C sink over the next 100 years, but rather to highlight key areas of model uncertainty requiring further experimental clarification. Here we show that improved understanding of the processes underlying soil N immobilization is essential if we are to predict the future course of the forest carbon sink.