Background: Plant traits related to nutrition have an influential role on tree growth, tree production and nutrient cycling. Therefore, the breeding program should consider the genetics of the traits. However, the measurement methods could seriously affect the progress of breeding selection program. In this study, we tested the ability of spectroscopy to quantify the specific leaf nutrition traits including Anthocyanins (ANTH), flavonoids (FLAV) and Nitrogen balance index (NBI), and estimated the genetic variation of these leaf traits based on the spectroscopic predicted data. Live fresh leaves of Sassafras tzumu were selected for spectral collection, after which concentrations of ANTH, FLAV and NBI were analyzed by standard analytical methods. Partial least squares regression (PLSR), five spectra pre-processing methods, and four variable selection algorisms were conducted for the optimal prediction model selection. Each trait model was simulated 200 times for error estimation. Results: The Standard Normal Variate (SNV) to the ANTH model and 1st derivatives to the FLAV and NBI models, combined with significant Multivariate Correlation (sMC) algorithm variable selection are finally regarded as the best performance model. The ANTH model produced the highest accuracy of prediction with a mean R2 of 0.72 and mean RMSE of 0.10 %, followed by FLAV and NBI model (mean R2 =0.58, mean RMSE = 0.11 % and mean R2 =0.44, mean RMSE = 0.04 %). High heritability was found of ANTH FLAV and NBI with h2 of 0.78, 0.58 and 0.61 respectively. It shows that it is benefitting and possible of breeding selection for the improvement of leaf nutrition traits. Conclusions: Spectroscopy can successfully characterize the leaf nutrition traits in living tree leaves and the ability to simultaneous multiple plant traits provides a promising and high-throughput tool for the quick analysis of large size samples and serves for genetic breeding program.