Applicability of near-infrared spectroscopy for measurement of amino and fatty acid composition in whole soybeans was the main subject of three research papers included in this dissertation. The effects of type of spectrometer, calibration method, and data preprocessing techniques were also investigated.;Validation of developed amino acid calibration models resulted in r2 values ranging from 0.04 (tryptophan) to 0.91 (leucine and lysine). Most of the models were usable for research purposes and sample screening, however, no sufficient correlation was found between spectral data and concentrations of cysteine and tryptophan. It was established that the variation in predictive ability of equations was determined by how a certain amino acid correlated to reference protein. Comparison of calibration methods demonstrated that (1) performance of partial least squares and support vector machines regressions was significantly better than that of artificial neural networks, and (2) choice of preferred modeling method was spectrometerdependent.;Validation of developed fatty acid calibration equations demonstrated that (1)equations for total saturates had the highest predictive ability ( r2 = 0.91--0.94) and were usable for quality assurance applications, (2) palmitic acid models (r2 = 0.80--0.84) were usable for certain research applications, and (3) equations for stearic (r2 = 0.49--0.68), oleic (r2 = 0.76--0.81), linoleic ( r2 = 0.73--0.76), and linolenic (r 2 = 0.67--0.74) acids could be used for sample screening. The results also showed that support vector machines models produced significantly more accurate predictions than those developed with partial least squares regression. Neural networks calibrations were not significantly different from the other two methods. Reduction of number of calibration samples reduced predictive ability of all types of equations, however the rate of performance degradation of support vector machines models was the lowest.;The third study compared applicability of global and local implementations of principal component analysis compression to near-infrared calibration problems solved with the neural networks regression. Two lysine data sets were used for development of control and experimental calibrations. The results demonstrated that local principal component compression could significantly outperform its traditional global counterpart.