The potential of near-infrared hyperspectral imaging, as a rapid and nondestructive technique with the spectral wavelength range of 899~1 694 nm, was conducted to predict moisture content (MC) in Atlantic salmon fillets. Altogether 100 fish fillets cutting out from different parts of 5 whole fillets were collected for hyperspectral image scanning. Mean spectral data were extracted automatically from the region of interest (ROI) of Atlantic salmon fillet surface. In order to reduce high dimensionality of hyperspectral images, successive projections algorithm (SPA) was performed to select optimal wavelengths for detection of MC in Atlantic salmon fillets. Partial least square (PLS) was carried out for the detection of MC in Atlantic salmon fillets based on spectral. The results showed that SPA-PLS achieved satisfactory result with R2 of 0.913 and 0.904, RMSEP of 0.965% and 1.169% for both calibration and prediction sets respectively. Then SPA-PLS models were built pixel-wise to the hyperspectral images of the prediction samples to produce chemical images for visualizing MC distribution. The results demonstrated that the potential of hyperspectral imaging technique to predict MC distribution in salmon fillets.