The increasing number of breast cancer in recent years has attracted numerous researchers' attention. Several techniques of Computer Aided Diagnosis System have been proposed as alternative solutions to diagnose breast cancer. The flaw of simply using the naked eye to see the differences between normal and with cancer mammogram images makes the texture analysis play an important role in classifying breast cancer. In this study, the results of the classification were compared using various methods of texture analysis in extracting a feature of the mammogram image. Some texture analysis methods, including first order, which consist of GLCM, GLRLM, and GLDM, have successfully extracted features based on their characteristics. The statistical features of these methods are used as input for the ECOC SVM classification, which three kernel comparisons; linear, RBF, and polynomial, build the classification. The results show that the best kernel is polynomial kernels with statistical features built by GLRLM with 93.9757% accuracy value.