Tuberculosis is one of the causes of human death. The results of the x-ray examination of tuberculosis diagnosis can be used as an object in the feature extraction process which is a stage in extracting the characteristics of the object contained in an image of a diagnosis of tuberculosis. In this study used first-order statistic (histogram), first-order Gray-Level Co-occurrence Matrix (GLCM) feature extraction methods, as well as the Principle Component Analysis (PCA). Data research digital x-ray tuberculosis patients from Dr. Sardjito Yogyakarta as 33 patients in 2012. Each 6 normal PA (Postero-anterior), 19 abnormal PA, 4 normal AP (Antero-Posterior), and 4 abnormal AP. This study aims to find the best characteristics contained in the x-ray image of tuberculosis diagnosis using statistical texture analysis obtained from features found in feature extraction methods. Identified features: variance, standard deviation, skewness, kurtosis, contrast and energy. Classification uses 33 test data are built using the Multi Layer Perceptron (MLP) method, while the output is a normal and abnormal image. The results showed that the accuracy classification used Histogram (81,81%), GLCM (96,96%), PCA (81,82%), and combination GLCM Histogram (100%).