According to the characteristics of near infrared spectral(NIR)data, a new tactic called stability competitive adaptive reweighted sampling (SCARS) is employed to select characteristic wavelength variables of NIR data to build PLS model. This method is based on the stability of variables in PLS model. SCARS algorithm consists of a number of loops. In each loop, the stability of each corresponding variable is computed at first. Then enforced wavelength selection and adaptive reweighted sampling (ARS) is used to select important variables according to the stability of variables. The selected variables are kept as a variable subset and further used in the next loop. After the running of all loops, a number of subsets of variables are obtained and root mean squared error of cross validation (RMSECV) of PLS models is computed. The subset of variables with the lowest RMSECV is considered as the optimal variable subset. Validated by NIR data set of protein fodder solid-state fermentation process, the SCARS-PLS prediction model is better than PLS models based on wavelengths selected by competitive adaptive reweighted sampling (CARS) and Monte Carlo uninformative variable elimination (MC-UVE) methods. As a result, twenty one wavelength variables are selected by SCARS method to build the PLS prediction model with the predicted root mean square error (RMSEP) valued at 0.054 3 and correlation coefficient (Rp) 0.990 8. The results show that SCARS tactic can efficiently improve the accuracy and stability of NIR wavelength variables selection and optimize the precision of prediction model in solid-state fermentation process. The SCARS method has a certain application value.