The aim of this study was to compare visible-near-infrared spectroscopy (VIS/NIRS) models developed from fresh or freeze-dried samples for predicting the fatty acid (FA) composition of beef samples. The hypothesis tested is that the removal of water from samples could improve the VIS/NIRS model performance. A total of 454 beef samples obtained from different bovine muscles were used. No significant differences were found in the performance of VIS/NIRS models developed from fresh or freeze-dried samples for predicting both major individual FAs and families of FAs and for some FAs (16:0, 18:0, 18:1 n-9, 18:2 n-6, 20:4 n-6, 22:5 n-3, 22:6 n-3, saturated, mono-unsaturated FA, and total n-3 long chain poly-unsaturated FAs (PUFA)). In contrast, the standard error of predictions for total PUFAs, total n-3 PUFAs, total conjugated linoleic acid, 20:5 n-3, and 18:3 n-3 were improved (by 21% on average; P < .05) in freeze-dried samples compared with fresh samples.