The objective of this study is to compare different calibration models that could be used in the analysis of protein, oil, carbohydrate and ash contents in maize flour by NIRS. A total of 138 samples were used from 115 hybrids and 23 inbreds in the study as material. Based on reference analysis results, different estimation models were developed using Partial Least Squares Regression (PLSR) and Multiple Linear Regression (MLR) methods. Validation procedure of these models (n=110) were accomplished using samples from different genotypes (n=28). In both of the developed models, the highest accuracy was attained for protein content (r=0.990 for MLR and r=0.987 for PLSR). For the other traits analyzed, although MLR model yielded better results based on mathematical evaluations (r(MLR)=0.801, r(PLSR)=0.755 for carbohydrate, r(MLR)=0.823, r(PLSR)=0.723 for oil, r(MLR)=0.926 and r(PLSR)=0.810 for ash), external validation suggested PLSR model provide a lower error rate than MLR. Results suggested that protein content could be successfully estimated, whereas, for some other traits, such as carbohydrate and oil ratios, it seems that there is still need for more studies before getting accurate measurements using NIR methods. Profile analysis regarding the wavelengths potent in the models showed that the estimation power declined when the regression coefficients of the wavelengths included in the model were low. Among the analyzed traits, ash and oil contents seemed to be related with more spectral regions within the scanned spectra than protein and carbohydrate.