Near infrared reflectance spectroscopy (NIRS) is widely used to assess the nutrient composition of forages. In forage, the leaf to stem ratio of alfalfa greatly affects its forage quality, with a high ratio of leaf indicated as high quality. This study aimed to evaluate the predictability of the alfalfa leaf-to-stem ratio and feed value using NIRS. Alfalfa hay was manually separated into leaves and stems by hand and the analysis samples were then made in the controlled range between 0 and 100%. Calibration models (n=320) were developed using modified partial least squares regression (MPLS) based on cross-validation. The optimal calibrations were selected based on the highest coefficients of determination in cross-validation (R2) and the lowest standard error of cross-validation (SECV). The prediction accuracy for the leaf-to-stem ratio (SECV, 5.95 vs. 5.71%; R2, 0.91 vs. 0.91) in alfalfa hay was comparable. For leaves, the standard error of calibration (SEC) was 4.94% (R2=0.94), and for stems, it was 4.81% (R2=0.94). The leaves and stems of the SEC were 4.94% (R2=0.94) and 4.81% (R2=0.94), respectively. The prediction accuracy for feed value, based on the leaf-to-stem ratio, predicted SECV values of 0.92% (R2=0.88) for crude protein (CP) content, 1.92% (R2=0.91) for neutral detergent fiber (NDF) content, 1.36% (R2=0.91) for total digestibility nutrient (TDN) content, and 9.86 (R2=0.81) for relative feed value (RFV). The results of this study demonstrate the potential of the NIRS method as a reliable tool for predicting the leaf-to-stem ratio of alfalfa hay, and show available techniques for routine feed value evaluation.