Eddy covariance measurements of net ecosystem carbon dioxide (CO2) exchange (NEE) were taken at an ombrotrophic bog near Ottawa, Canada from 1 June 1998 to 31 May 2002. Temperatures during this period were above normal except for 2000 and precipitation was near normal in 1998 and 1999, above normal in 2000, and well below normal in 2001. Growing period maximum daytime uptake (-0.45 mg CO2 m-2 s-1) was similar in all years and nighttime maximum respiration was typically near 0.20 mg CO2 m-2 s-1, however, larger values were recorded during very dry conditions in the fourth year of study. Winter CO2 flux was considerably smaller than in summer, but persistent, resulting in significant accumulated losses (119--132 g CO2 m-2 period-1). This loss was equivalent to between 30 and 70% of the net CO2 uptake during the growing season. During the first 3 years of study, the bog was an annual sink for CO2 (~-260 g CO2 m-2 yr-1). In the fourth year, with the dry summer, however, annual NEE was only -34 g CO2 m-2 yr-1, which is not significantly different from zero. We examined the performance of a peatland carbon simulator (PCARS) model against the tower measurements of NEE and derived ecosystem respiration (ER) and photosynthesis (PSN). PCARS ER and PSN were highly correlated with tower-derived fluxes, but the model consistently overestimated both ER and PSN, with slightly poorer comparisons in the dry year. As a result of both component fluxes being overestimated, PCARS simulated the tower NEE reasonably well. Simulated decomposition and autotrophic respiration contributed about equal proportions to ER. Shrubs accounted for the greatest proportion of PSN (85%); moss PSN declined to near zero during the summer period due to surface drying.