This experiment was carried out to find suitable sample type for a more accurate prediction and non-destructive way in the application of NIRS technique for estimation amino acid composition of soybean by comparing three different sample types, single seed, whole seeds, and milled seeds’ powder. The coefficient of determination in calibration and 1-VR of cross-validation for 17 amino acids analyzed by NIR using milled powder were highest, followed by single seed, and then whole seeds were the lowest. The R2 coefficients of determination in calibration for single bean perimeters were higher than those of whole beans and showed higher R2 coefficient than 0.8 with the exception of seven amino acids. On the other hand, calibration equation development for only glutamic acid analyzed NIRS data scanned with whole seeds showed higher R2 coefficient than 0.8. Eleven different amino acids, such as aspartic acid, threonine, serine, glutamine, glycine, alanine, valine, isoleucine, leucine, arginine and proline, of soybean seeds in the powder had higher R2 coefficient of determination in calibration than 0.8 and could be estimated for the most accurate prediction. However, judging from the results of single bean samples, in consideration on NIR application for more nondestructive and faster prediction, these amino acids using single bean sample could be estimated without grinding the seeds.