This research aimed to develop regression models that estimate starch concentration in saffron (Crocus sativus L.) corms from hyperspectral light scattering images. Light scattering images were captured from corms at wavelengths from 650 nm to 1,000 nm in 5 nm intervals. Light decay curves were measured, and the integrated value of the curve at each wavelength (S l ) was calculated by image processing algorithms. The starch concentration in each captured corm was also measured using the phenol sulfuric acid colorimetric method for validation. A principal component regression method was applied to develop regression models in order to estimate the starch concentration from the S l spectra. The results indicated that the estimation accuracy was high, and this model had practical use based on the ratio of performance to deviation (RPD) criterion (R cal 2 ϭ 0.913, standard error of calibration (SEC) ϭ 1.33 w.b., R val ϭ 0.932, standard error of validation (SEP) ϭ 1.54 w.b., RPD ϭ 2.81). The S l was found to be negatively correlated with starch concentration since light scattering increased considerably as starch concentration increased.