Rare earths are important strategic reserve resources, which are widely used in industry, agriculture, military, etc. However, high-precision detection of trace rare earth elements based on the combination of laser⁃induced breakdown spectroscopy (LIBS) technology and conventional unit calibration methods still remains challenging. A multivariate linear calibration model is established using preferred multispectral lines, replacing the conventional unit calibration methods, to improve the accuracy of LIBS detection for four rare earth trace elements (La, Gd, Y, Yb). On the basis of optimizing experimental parameters (defocus distance, acquisition delay, and laser pulse energy), a multivariate linear calibration model for the four elements was established, and the root mean square errors (RMSE) of the calibration model under one to three characteristic spectral lines are compared based on validation samples. The results show that the RMSE of La, Gd, Y, and Yb decreased from 0.00341%, 0.00727%, 0.00662%, and 0.00653% in the single linear calibration fitting to 0.00203%, 0.00707%, 0.00586%, and 0.00356% in the ternary linear calibration fitting model. This result indicates that the use of LIBS multivariate linear calibration model can significantly reduce prediction error of rare earth elements, and the detection accuracy of rare earth elements can be effectively improved.