<p>In laser-induced breakdown spectroscopy (LIBS), the quantitation accuracy with traditional standard calibration methods is often unsatisfactory, mainly because of the matrix effect. In this work, a novel method named specific-intensity clustering (SIC) was proposed to mitigate the influence of the matrix effect. The specific-intensity, which is defined as the ratio of the spectrum intensity to element concentration, is considered an essential characteristic of a sample. These specific-intensity features facilitate the effective clustering of calibration samples, thereby enabling the establishment of multiple, more robust, calibration curves. By combining the clustered calibration samples with machine-learning algorithms, a classification model was established. Unknown samples can be classified by the classification model and predicted by the proper calibration curve. The feasibility of the proposed SIC method was verified through the detection of CaO, TFe₂O₃, and Li in rock samples. Compared with traditional standard calibration methods, the SIC method significantly reduced the average relative errors of the prediction set (AREPs) for the detected CaO, TFe₂O₃, and Li elements in rocks, with the AREPs decreasing from 25.292, 14.114, and 14.12% to 6.165, 5.481, and 5.137%, respectively. This work demonstrates that the SIC method has excellent potential for reducing the influence of the matrix effect in LIBS quantitation.</p>