Tenderness is a critical quality characteristic of salmon fillets and Warner-Bratzler shear force (WBSF) is a widely used objective indicator for tenderness evaluation of salmon fillets. This research studied rapid and non-destructive prediction of tenderness in fresh farmed salmon fillets using visible and near-infrared (Vis-NIR) hyperspectral imaging. Hyperspectral images of tested fillets with different tenderness levels were acquired and their spectral features were extracted in 400-1720 nm. Two calibration algorithms, namely partial least squares regression (PLSR) and least-square support vector machine (LS-SVM) analysis, were used to correlate the extracted spectra of salmon samples with the reference tenderness values estimated by WBSF method. Optimal wavelength selection was carried out based on full range spectra with two methods, regression coefficients (RC) from PLSR analysis and successful projections algorithm (SPA). The best set of optimum wavelengths was determined as the one containing four wavelengths (555, 605, 705 and 930 nm) selected by SPA. These four optimum wavelengths were then used to build an optimised SPA-LS-SVM prediction model, reaching the best result with a correlation coefficient (r(p)) of 0.905 and root mean square error estimated by prediction (RMSEP) of 1.089. At last, an image processing algorithm was developed to transfer the SPA-LS-SVM model to each pixel in salmon fillets for visualising their WBSF distribution. The overall results of this study reveal the capability of hyperspectral imaging as a fast and non-invasive technique to quantitatively predict tenderness of salmon fillets with a good performance. (C) 2013 Elsevier Ltd. All rights reserved.