Aiming at the problem of periodic arch protrusion in laser-induced breakdown spectroscopy (LIBS) data, an improved baseline correction algorithm was proposed. By adjusting the local penalty coefficients, the algorithm which was based on penalized least square method could not only fit the periodic arch protrusion, but also reduce the effect of spectral lines intensity on fitting baseline. Compared with other baseline correction methods, the root-mean-square error (RMSE) of the proposed algorithm was smaller when fitting the simulation baseline, and the correlation coefficients of the calibration curve based on spectral data obtained by the proposed algorithm reached 0.997 2. The results show that compared with the existing baseline removal method, the proposed algorithm can better retain the effective information of the spectral data when the echelle grating spectrometer collecting the baseline of LIBS.