Identification of susceptible host genes plays a key role in the treatment and prevention of tuberculosis. So far, only a few genes have been confirmed to be related to it. Based on the gene chip dataset GSE54992 of peripheral blood mononuclear cells of tuberculosis patients, this paper selects the differentially expressed genes between normal samples and active tuberculosis samples by combing information prior Bayesian test and linear model empirical Bayesian method. 319 differentially expressed genes are recognized by both methods. These genes are modeled by using naive Bayes classifier based on the independent verification set GSE83456 where a high classification accuracy is obtained. Finally, the molecular mechanism of tuberculosis is analyzed from the biological point of view through GO function enrichment and KEGG pathway analysis. This study highlights the important roles of the three Bayesian methods in gene data analysis. It provides a new comprehensive strategy for exploring specific biomarkers of tuberculosis, and shows the important clues for the prevention, diagnosis and treatment of tuberculosis.