In the present study, NIRS was applied to nondestructive and rapid measurement of firmness and surface color of pear. In order to improve the prediction precision and eliminate the influence of uninformative variables on model robustness, Monte Carlo uninformative variables elimination (MC-UVE) and Monte Carlo uninformative variables elimination based on wavelet transform (WT-MC-UVE) methods were proposed for variable selection in firmness and surface color NIR spectral modeling. Results show that WT-MC-UVE can reduce the modeling variables from 1 451 to 210, and get similar prediction results for firmness. WT-MC-UVE improved the prediction precision for surface color, the root mean square error of prediction (RMSEP) and calibration variables were reduced from 1.06 and 1 451 to 0.90 and 220 respectively, and the correlation coefficient (r) was improved from 0.975 to 0.981. The proposed method is able to select important wavelength from the NIR spectra, and makes the prediction more robust and accurate in quantitative analysis of firmness and surface color.