INTRODUCTION The quantity and localisation of lipids in fish meat is linked to its quality (1). Measurement of the distribution of lipids inside fish is thus of great interest. This can be achieved by MRI using a T1-weighted SE since the signal depends linearly on the fat content (2). However, RF spatial inhomogeneities may lower the accuracy of fat measurement and correction is needed (3) ; the first purpose of this work is to test the effect of image correction on fat content estimation. Moreover, it is convenient to make measurement on frozen-thawed fish in order to postpone analysis. As freezing-thawing process may modify fish meat structures the second purpose of this work is to test its influence on the MRI signal. MATERIALS AND METHODS Measurements were performed on twenty-seven half-slices of trout. The slices were frozen at -20°C and thawed 24h at 4°C. MRI acquisitions were done at 4°C using a SE, with five combinations of flip angle and TR to enable the correction (4). The FOV was 200 mm *200 mm which was sufficient to put in three half-slices together. Subcutaneous fat was removed and the remainder was analysed using NMR (5). Subcutaneous fat was manually removed in the images and images were corrected. The error of calibration (RMSE) of NMR using MRI was computed. RESULTS AND DISCUSSION The mean fat quantity was 9.6 % with a standard deviation of 1.6 %. For fresh fish, the RMSE was 1.25 % and 0.76 % respectively without and with image correction. For frozen and thawed fish the RMSE was 1.18 % and 0.81 % respectively without and with image correction correction. Two major conclusions could be drawned from this study : thawing did not affect significantly the results of prediction and correction of the images improved the results.