The combination of the near infrared (NIR) and Fourier-transform infrared (FTIR) absorbance spectra (1100-2500nm and 4000-600cm(-1)) of 100 cocoa powder samples was used to build calibration models for the determination of the content of fat, nitrogen, and moisture. The samples that comprised the dataset had an average composition of 13.51% of fat, 3.77% nitrogen, and 3.98% moisture. The fat content ranged from 2.42 to 22.00%, the nitrogen from 0.88 to 4.48%, and moisture from 1.60 to 7.80%. For NIR, the relative root mean square error of cross-validation (RMSECV) was 7.0% (R(2)=0.96) for fat, 1.7% (R(2)=0.98) for nitrogen, and 5.2% (R(2)=0.94) for moisture. For FTIR, the relative RMSECV was 10.4% (R(2)=0.94) for fat and 3.9% (R(2)=0.95) for nitrogen. However, for moisture, it was not possible to build a calibration model with suitable predictability. The combination of the NIR and FTIR domains (data fusion) by outer product analysis PLS1 allowed to predict these parameters and to characterise frequencies in one domain based on the information of the other domain. This work allows to conclude that the second derivative of NIR is the recommended procedure to quantify fat, nitrogen, and moisture content in cocoa powders by infrared spectroscopy.