The PnET-Day model was independently parameterized to compare with estimated eddy covariance gross ecosystem production (GEP; gCm⁻² day⁻¹) in a mature mixed hardwood and a mature red pine (Pinus resinosa) forest in Northern Wisconsin during the growing season of 2002 and 2003. The mature hardwood forest was dominated by Populus tremuloides, Populus grandidentata, Betula papyrifera, Quercus rubra, Acer rubrum, and Acer saccharum. We evaluated the model's capability to predict the seasonal and interannual dynamics of GEP and explored the sources of discrepancy between PnET-Day and eddy covariance GEP estimates. GEP was directly estimated from the two eddy-flux towers, one for each forest type, during 2002 and 2003. PnET-Day growing season GEP for the mature hardwood forest was 12% higher in 2002 and 12% lower in 2003 than eddy covariance GEP estimates, while the modeled growing season GEP of the mature red pine forest was overestimated by 43 and 32% compared to eddy covariance GEP in 2002 and 2003, respectively. The disagreement between the two methods was attributed to year-to-year variability in foliage biomass and foliar nitrogen (N) in the mature hardwood forest and to high foliage biomass and specific leaf weight in the mature red pine forest (>50% larger than red pine in Harvard Forest where the model was developed and validated). The difference between PnET-Day and eddy covariance GEP estimates was greatest in May in the hardwood forest, primarily due to the discrepancy between the true and parameterized foliage onset day. Our results suggest that improved prediction of foliage onset is necessary to improve PnET-Day estimation of GEP in a hardwood forest.