A total of 445 samples with great variability in amino acid contents were harvested for different seasons in different regions for developing calibration equations of amino acid content in cottonseeds. The spectral data of cotton kernel powder was processed using the first derivative mathematical treatment combined with SNV and de-trend, as well as modified partial least squares (MPLS) regression method. The chemometric models for 17 amino acids present in cottonseed were developed, and 12 of them were excellent for the determination of related amino acids, namely asparagic acid, threonine, glutamic acid, glycine, alanine, valine, isoleucine, leucine, phenylalanine, lysine, histidine, and arginine, with RPDc of 3.735~7.132 and determination coefficient (r2) of 0.910~0.979 in external validation. For those 12 amino acids, their values predicted by NIRS are comparable to those obtained by the chemical method with good accuracy. The RPDc of serine, methionine, tyrosine and proline were 2.205~2.814, and their determination coefficient (r2) were 0.800~0.830 in external validation. For those 4 amino acids, the values from NIRS are not so accurate as chemical analysis, but could be used in sample screening in cotton breeding program. While the equation for cystine was useless as its RPDc was only 1.358, which was not suitable for estimating its content in cottonseeds.