The radiance recorded at a sensor is not fully a representative of Earth surface section features but is altered by atmosphere. In this study, we evaluated three atmospheric correction (AC) methods (a typical empirical modeling method, a radiative transfer modeling approach, and a combination of the both methods) in identifying urban tree species/groups with high-resolution WorldView-2 (WV2) imagery in the City of Tampa, FL, USA. We tested whether AC methods were necessary in urban tree species discrimination. In situ spectral measurements were taken from tops of tree canopy and tree crowns were delineated from WV2 imagery. Two-sample t-tests, repeated measures ANOVA (RANOVA) tests, linear discriminant analysis (LDA), and classification and regression trees (CART) classifiers were used to test the spectral difference between in situ spectra and atmospherically corrected image spectra and to discriminate urban tree species/groups. The experimental results demonstrate that 1) the empirical line-based AC methods were relatively more effective than a radiative transfer-based AC model to atmospherically correct the image data, due to lacking accurate and reliable atmospheric parameters to run the radiative transfer model and 2) the AC processing to WV2 imagery was unnecessary in identifying seven tree species/groups in this particular case, most likely because the WV2 image data used in this analysis were acquired on a single date and covered a relatively small area (303 km2). The study results also indicate that compared with a nonparametric classifier CART, the parametric classifier LDA produced higher overall accuracy (55% vs. 48%) for identifying the seven species/groups.