Based on the analyses of 43 elements in 16 samples of the raw coal and feed coal collected from the northern Ordos basin and Shanxi Province, the modes of occurrence of these elements were studied using the method of cluster analysis and factor analysis, and the cleaning potential of the hazardous elements relatively enriched in the coals was discussed by analyzing six samples of the cleaned coal from the coal-washing plants and coal cleaning simulation experiments. The results shows that the elements Br and Ba show a strong affinity to the organic matter, Cs, Cd, Pb, Zn and Hg partly to the organic matter, and the other trace elements are mainly associated with the mineral matter. Cs, Mo, P, Pb, Zn and S have positive correlations with the two principal factors, reflecting the complexity of their modes of occurrence. Some elements that were thought to show a faint relationship (Be with S and Sb with carbonates) in other rocks are found to have a strong interrelation in the coals. Clay minerals (mainly kaolinite) dominate in the coals, and Ta, Th, Ti, Sc, REE, Hf, U, Se, W, V, Nb, Mo, Al, P, Cr, Pb and Zn are distributed mostly in kaolinite, while K, Rb, Cs, and Na have much to do with illite. Conventional cleaning can reduce the concentrations of most hazardous elements in various degrees. The hazardous elements S, As, Sb, Se, Mo, Pb, Cd and Hg relatively enriched in some coals from the area studied have a relatively high potential of environmental risks. However, by physical coal cleaning processes, more than 60% of As and Hg were removed, showing a high degree of removal, more than 30% of Sb, as well as S, Pb and Cd partly associated with the inorganic matter were removed. Se and Mo showing a relatively low degree of removal could be further removed by deep crushing of the coal during physical cleaning processes, and the concentrations of S, Pb, Cd and Hg with a partial association with the organic matter could be decreased in such ways as the coal blending. Cluster analysis together with factor analysis is a rapid and effective way to deduce the mode of occurrence of an element from bulk samples, and the removability data of most hazardous elements are basically consistent with their modes of occurrence suggested, which indicates that the statistical analysis could predict the cleaning potential of hazardous elements during the physical coal cleaning.