There is a need to speed up proximate analyses of quinoa by means of inexpensive, rapid and accurate methods such as those based on spectroscopy. Thus, the aim of this study was to assess the feasibility of using Near-Infrared Transmission (NIT) spectroscopy for precise and low-cost determinations of protein and fat contents of quinoa grains. NIT spectrum of Peruvian-origin quinoa grains of 73 different accessions were obtained while total protein and fat contents were determined by AOAC methods. To minimise the effects of changes in the baseline, spectra were pre-processed by separately applying a Savitzky-Golay algorithm first (SG1), second derivative (SG2), standard normal variate (SNV), multiplicative scatter correction (MSC) and Detrend (DT) filters. In addition, MSC filters were also applied in combination with SG1, SG2, SNV and DT. Best predictions for fat were obtained by extracting 16-18 partial least square (PLS) components from SNV, DT and SG2-treated spectra (mean square error of prediction, RMSEP=0.432-0.451; R2=0.728-0.786) while best predictions for protein were attained by extracting 14-16 PLS components from DT, SG2+MSC and DT+MSC-treated spectra (RMSEP=0.545-598; R2=0.833-0.862). Crossvalidation run for ten-observation segments showed that model accuracy was slightly better for fat than for protein.