The objective of the study was to evaluate the ability of near infrared (NIR) spectroscopy to predict pork water-holding capacity (WHC) in relation to different methods of WHC determination. The study comprised 228 longissimus dorsi muscle samples (147 for calibration, 81 for external validation). Calibration models were developed using WinISI II on visible (lambda = 400-1100 nm), NIR (lambda = 1100-2500 nm) or the whole spectral range (lambda = 400-2500 nm). Models' quality was assessed by means of the coefficient of determination of cross-validation (R(cv)(2)) or prediction (R(p)(2)) and the error of cross-validation (se(cv)) or prediction (se(p)). The R(cv)(2) ranged from 0.28 to 0.62, while se(cv) was estimated between 1.0% and 1.4% for EZ drip loss (meat juice container procedure), 2.3% and 2.5% for cooking loss, 1.9% and 2.3% for centrifuge force and 0.9% and 1.0% for tray drip loss, and was similar to the repeatability (s(r)) of the method (1.1%, 1.8%, 2.0%, 0.4%, respectively). The results of the external validation confirmed those obtained by cross-validation (R(p)(2) = 0.22-0.44; se(p) = 1.2-1.3%, 2.2-2.4%, 2.1-2.3% and 1.1% for EZ drip loss, cooking loss, centrifuge force and tray drip loss, respectively). The best calibration results were obtained for EZ drip loss (R(cv)(2) = 0.62, se(cv) = 1.0%) and tray drip loss (R(cv)(2) = 0.49. se(cv) = 0.9%) on visible spectral range. NIR spectroscopy offers great advantages for the on-line application, however, its prediction ability is limited by the reference methods and this should be taken into account. (C) 2010 Elsevier Ltd. All rights reserved.