Many studies have successfully applied near infrared (NIR) spectroscopy to estimate important wood traits. Some of them have reported the effects of wood surfaces on NIR spectra information and their influence on the performance of the predictive model. However, limited information is available concerning the effect of sample preparation on the model performance to estimate chemical properties in Eucalyptus wood. Hence, the aim of this study was to investigate the influence of the milling procedure, particle size, and quality of the solid wood surface on the performance of the partial least squares regression to predict chemical properties of Eucalyptus urophylla wood by NIR spectroscopy. Adequate models were built to predict klason lignin content, acid-soluble lignin content, and syringyl-to-guaiacyl ratio in Eucalyptus urophylla wood. Sample preparation strongly influences the ratio of performance to deviation (RPD) of these predictive models. The effect of the sample presentation (solid or milled wood) was stronger than the effect of the particle size difference between thin and thick powder. The best calibrations were developed using NIR spectra measured on wood powder (RDP values from 1.99 to 2.97), but satisfactory calibrations were developed from NIR spectra measured on solid samples (RPD values from 1.68 to 2.16).