Forest managers are increasingly called upon to provide long-term predictions of forest development. The dynamics of regeneration establishment, survival and subsequent recruitment of established seedlings to larger size classes is a critical component of these forecasts, yet remains a weak link in available models. To test the reliability of FVS-NE for simulating sapling (stems ≥ 0.5 in dbh) recruitment dynamics in stands subject to repeated partial harvests, we compared model predictions with long-term observations (n = 729 plots/5-yr interval combinations) from the Penobscot Experimental Forest (PEF) in central Maine. Two different parameterizations of FVS-NE were tested; the currently available production code and a yet to be released beta version that contains a number of structural changes. Because neither parameterization has a full-establishment model, regeneration composition and densities were from the research plots. Our analyses indicated that predicted rates of sapling recruitment were biased according to both models, averaging 47 percent (production) and 206 percent (beta) of the observed rate at the PEF (1.71 ± 0.25 ft2/ac/5-yr). Mortality rates among the newly recruited saplings were overestimated by both models, and species composition of the survivors did not closely match the observations. Correlation analysis on the residuals from the beta version pointed to a strong link between the overestimation of stems recruited to the sapling size class and the density of large regeneration input to the model. Limiting the density of regeneration entering the simulation to ≤1,800 TPA (~2 stems/milacre plot) largely eliminated the prediction bias, yet only modestly improved model accuracy (R2 0.398 vs. 0.341).