Within the framework of a project requiring an important number of measures of lipid content in chicken breast meat, near-infrared spectrometry (NIRS) was tested in order to avoid the use of the reference extraction method requiring long analytical process. Two methods were compared: direct spectrum acquisition with a fibre optic probe on breast muscle, 24 h after slaughter and just after the cutting and boning process (spectrometer ASD, Labspec Pro) or later in the laboratory on ground breast meat (spectrometer Büchi). The direct spectral acquisition was realised on 877 fillets of broilers issued from different slaughter plants. The laboratory spectral acquisition was realised on the same 877 fillets plus 457 other samples also issued from different slaughter plants. Calibration equations established with ground meat were more precise. SECV (standard error of cross-validation) was 0.24% (R2=0.83) for lipids, which is equivalent to the repeatability of reference measurements (0.20%). The direct measurements on breast muscle led to a less precise calibration equation (lipids: SECV=0.37%, R2=0.68). In conclusion, the prediction of lipid content with NIRS provided satisfactory data when spectra were realized in the laboratory on ground breast meat. Direct measurement was less precise, but allowed a useful in situ estimation of the chemical composition.