The miscellaneous materials, including gangue, plastic, and wood commonly present in coal. These miscellaneous materials affect the reliability of coal analysis using laser-induced breakdown spectroscopy in power plants, but have significantly distinct spectral characteristics from coal. Hence, this paper proposes a step-by-step classification method to screening the false spectra of miscellaneous materials. The first step aims to identify the plastic and wood spectra by determining the existence of specific characteristic spectral lines using the standard deviation (SD) values. The spectral lines Si Ι 288.16 nm with the SD value of more than 850 counts and Li Ι 670.78nm with SD value of more than 1750 counts were used as the distinguishing markers. The classification accuracy of first step was 100%. Due to the high similarity between gangue and coal, the second step utilized the random forest (RF) classification model to identify the gangue spectra. The number of trees and random variables in the RF model was optimized. The accuracy of classification model without and with the proposed step-by-step method was 98.30 and 99.96%, respectively. To assess the necessity of spectra classification, a set of calorific value analysis was performed by adding false spectra of different proportions, which were compared with analysis after removing the false spectra. The root mean square error of prediction (RMSEP) was 0.42 MJ kg-1 (after removing), compared with 0.50 MJ kg-1 (mixing with 10% gangue spectra), 0.56 MJ kg-1 (mixing with 20% gangue spectra) and 0.57 MJ kg-1 (mixing with 30% gangue spectra). The results demonstrated that the proposed step-by-step classification method could effectively identify the spectra of coal and miscellaneous materials and improve the accuracy of coal analysis.