In large-scale biodiversity surveys, the observed species-richness inevitably underestimates the real species richness of a certain geographic region. Nonparametric estimators may be feasible tools for predicting the potential tree/shrub species richness using data from large-scale forest surveys that gather reliable information. In light of the many nonparametric estimators available in the literature, we evaluated their statistical performance (bias and precision) in estimating the γ-diversity of different forest types by comparing their estimates with theoretical total richness values. In addition, we investigated their behaviors on simulated data regarding the spatial distribution of species among sample plots (patchiness). To accomplish these tasks, we used data from a wide-ranging systematic forest survey conducted at 418 sample plots on three subtropical forest types located in a Brazilian hotspot. We established theoretical real tree/shrub species-richness values for each forest type from a vast and carefully revised database for the neotropics. Due to the considerable amount of species with restricted geographic distribution, the incidence-based estimators were shown to be more suitable than the abundance-based estimators for estimating γ-diversity, presenting the smaller overall combination of performance metrics. The abundance-based estimators (Chao1 and ACE) were notedly sensitive to different degrees of patchiness, especially to patchy species distributions. On the other hand, Jackknife1 and the sample-based extrapolation were less influenced by patchiness. Regardless of the differences in species composition and structures among forest types, the Jackknife2 estimator presented the best overall performance and quite accurate estimates for all the cases.