This study investigated the application of near-infrared spectroscopy (NIRS) and multivariate calibration methods to the direct determination of the tannin content in Acacia mearnsii bark to improve control during the tannin extraction process. Eighty-nine bark samples were collected from the industrial plant of an extractive tannin industry. The NIR spectra were acquired using an FT-NIR spectrometer with an integrating sphere, an indium–gallium–arsenic detector in the range 7500–4000 cm⁻¹, a resolution of 16 cm⁻¹ and 32 scans divided in two different ways: (a) in natura samples (no sample processing); and (b) dried and milled samples. Partial least-squares models were developed and different strategies were investigated during pre-processing of the IR spectra. The results of the prediction were compared with those obtained using the reference methodology (NBR 11131) and gave values for the root mean square error of prediction between 2.11 and 2.42% for the dried and milled bark samples and between 2.31 and 2.54% for the in natura samples. These results show that NIRS combined with multivariate calibration methods may be used for the direct determination of the tannin content in Acacia mearnsii bark. The requirement for little sample preparation, a short analysis time, no reagent consumption and, consequently, no waste generation are the main positive characteristics of the proposed method.