Old-growth forests are often assumed to exhibit no net carbon assimilation over time periods of several years. This generalization has not been typically supported by the few whole-ecosystem, stand-scale eddy-covariance measurements of carbon dioxide exchange in old-growth forests. An eddy-flux tower installed in a >300-year-old hemlock–hardwood forest near the Sylvania Wilderness, Ottawa National Forest, MI, USA, observed a small annual carbon sink of CO2 of −72 ± 36 g C m−2 year−1 in 2002 and −147 ± 42 g C m−2 year−1 in 2003. This carbon sink was much smaller than carbon sinks of −438 ± 49 g C m−2 year−1 in 2002 and −490 ± 48 g C m−2 year−1 in 2003 observed by a nearby flux tower in a 70-year-old mature hardwood forest (Willow Creek, WI). The mature forest had vegetation similar to the old-growth site prior to European settlement. Both sites had slightly larger carbon sinks in 2003, which was a drier and cooler year than 2002. However, the difference in sink strength between the two years was smaller than the uncertainty in the results arising from missing and screened data. Both sites also had significant systematic errors due to non-representative fluxes during certain micrometeorological conditions, which required careful screening. The difference in sink strength between the two sites was driven mainly by greater ER at the old-growth site (965 ± 35 g C m−2 year−1 in 2002 and 883 ± 69 g C m−2 year−1 in 2003) compared to the mature site (668 ± 21 g C m−2 year−1 in 2002 and 703 ± 17 g C m−2 year−1 in 2003). GEP was lower at the old-growth site (1037 ± 47 g C m−2 year−1 in 2002 and 1030 ± 41 g C m−2 year−1 in 2003) compared to the mature site (1106 ± 47 g C m−2 year−1 in 2002 and 1192 ± 51 g C m−2 year−1 in 2003), especially in 2003. Observations also suggested that growing season ER had greater interannual variability at the old-growth site. These results imply that old-growth forests in the region may be carbon sinks, though these sinks are smaller than mature forests, mostly likely due to greater ER.