Feed with copper can accelerate the growth of pigs significantly, so it is common to find feed with excess copper content, but excess copper brings serious consequences. Laser induced breakdown spectroscopy (LIBS) technology is used to quantificationally analyze the copper in pig feed rapidly. Competitive adaptive reweighted sampling (CARS) algorithm screens 22 important wavelength variables which are associated with copper in pig feed with compression ratio of 1.1%. Finally, partial least squares (PLS) regression method is applied to establish the prediction model of copper content in pig feed based on the 22 important wavelength variables, and the copper content in prediction set pig feed samples is predicted. The results show that the CARS-PLS model has higher prediction accuracy and prediction ability than full spectrum-PLS model. The correlation coefficient, the root mean square error of cross validation and the relative error are 0.978, 19.25, 5.59%, respectively. CARS algorithm can effectively optimize the LIBS online detection model of copper in pig feed and improve the prediction accuracy.