Technical Abstract: Firmness of apple fruit is an important quality attribute, which varies in the same lot of fruit due to inherent biological variability, climatic condition, cultural practice, harvest time or maturity level, and postharvest handling and storage. This paper reports on using modified Lorentzian distribution (MLD) parameters of light scattering images, which were acquired by a compact multispectral imaging system over wavelengths between 650 nm and 1,000 nm, to predict apple fruit firmness. Optimal wavelengths for firmness prediction were determined with multi-linear regression and cross-validation. For Red Delicious, seven wavelengths (690, 770, 790, 810, 920, 980, and 1000 nm) were optimal, whereas for Golden Delicious, eight wavelengths (650, 690, 740, 750, 820, 880, 910, and 990 nm) were needed. The prediction models established based on the optimal wavelengths gave good firmness predictions with the correlation coefficient (r) of 0.82 and the standard error of validation (SEV) of 6.64 N for Red Delicious apples, and r = 0.81 and SEV = 6.58 N for Golden Delicious apples. These results were considerably better than those obtained with visible/near-infrared spectroscopy. The multispectral scattering imaging technique is potentially useful for sorting and grading apples and other fruits for firmness.