Eddy covariance observations of carbon dioxide fluxes from August 2009 to January 2013 are reported for an uneven-aged managed mixed-hardwood forest in central Ontario. The forest, dominated by sugar maple (Acer saccharum) is managed by selection-system silviculture with the last stand entry within the tower footprint in 1996/1997. Eddy covariance measurements were made from a 32 m high tower (about 10 m above the average canopy height). Annual NEP from 2010–2012 is estimated by gap-filling, and uncertainties due to quality control selection criteria, frequency and length of gaps, and choice of gap-filling algorithm are discussed. Results derived from a night-time u* threshold of 0.32 m s−1 imply 2010 was a significant carbon source (−4.15 Mg C ha−1 y−1) compared to 2011 and 2012 (−1.05 and −1.75 Mg C ha−1 y−1). Uncertainty due to quality control selection criteria is estimated to be in the range of 2.4 Mg C ha−1 y−1, while uncertainty due to gap-filling and gap-filling methodology is estimated to be in the range of 0.12 to 0.60 Mg C ha−1 y−1. With these in mind, CO2 fluxes in 2011 and 2012 are hardly distinguishable from carbon neutral. The high carbon loss in 2010 is attributed to an extreme heat event that occurred in late May and corresponded to the timing of peak leaf expansion that year, resulting in high leaf mortality and subsequently low leaf area index. Eddy covariance observations showed that gross ecosystem productivity during the growing season in 2010 was about 25% lower than in 2011 and 2012.