In order to predict husk content and rice production yield, 46 species of paddy were studied. After the near infrared diffuse reflectance spectroscopy, samples were preprocessed by wavelet denoising and Z-score normalizing. Kennard-Stone method was used to select the sample set. The competitive adaptive reweighted sampling was applied to screening the key wavelengths associated with the husk content and rice production yield from paddy. Finally, the prediction models of the husk content and rice yield were established with the multiple linear regression. The results indicated that 21 near infrared key wavelengths were associated with the husk content, and 28 key wavelengths were associated with rice yield. The R2 of two prediction models were 0.998 3 and 0.998 7. The root mean square error of calibrations were 0.112 9 and 0.982 1, and the relative deviation were 0.51% and 2.34%, respectively.