Tuberculosis, which is the oldest human disease with the highest mortality rates among infectious diseases, continues to be the world's attention. Previous methods to diagnose tuberculosis are tuberculin test, Sputum-smear microscopy and chest radiography. Unfortunately, these methods are time consuming and perform poorly. Furthermore, they require varied sensitivity, Mycobacterium tuberculosis bacilli alive, sputum which is difficult to obtain from children, trained personnel to avoid human error, and hence, high cost. Researchers keep developing accurate data mining methods for rapid Tuberculosis diagnosis to reduce the rate of growth of the world population of tuberculosis patients. This paper aims to provide state-of-the-art of data mining methods in diagnosing Tuberculosis using clinical symptoms as input parameters. First, it introduces tuberculosis and current methods used for tuberculosis diagnosis. Then it discusses techniques for preprocessing data and data mining methods for tuberculosis diagnosis currently used. The result shows that the most frequently used variables are sweating at night, more than 3 weeks of cough, fever, weight loss, age, and chest pain respectively. Support Vector Machine and Bayesian Network gave the highest accuracy compared to other methods.