Partial least squares projection analysis combined with uninformative variable elimination was used to select optimal wavelengths from apple hyperspectral scattering images.After this algorithm,the number of effectivewavelengths decreased to 26%. The selected effective wavelengths were set as inputs of partial least squares model. Root mean square error of prediction dropped from 6.00N to 5.73N and correlation coefficient increased a little. The result shows that to select effective wavelengths using the algorithm is feasible. In the parameter selection, there is not such as random defects. It is expected that the algorithm would provide an effective method foroptimal wavelengths selection using hyperspectral scattering image technique.