Abstract: Current clinical method often uses B-ultrasound for the diagnosis of fatty liver, which basically rely on the doctor observation. Therefore, automatic recognition of fatty liver from the B-ultrasound images has significance value of clinical application. According to the characteristics of fatty liver in B-ultrasound images, this paper proposed five useful features to formulate the new feature vector, which include the near-field echo density, the average gray ratio from the near to far field, as well as three texture features from the gray level co-occurrence matrix and the neighborhood gray-tone difference matrix such as angle second moment, entropy and busyness. These features are chosen carefully, which have good characteristic for distinguishing fatty liver. Based on the support vector machine classification of these feature vectors, the experiments show that the proposed method achieves a good recognition rate of normal liver and fatty liver, which is 91.7% and 98.7% respectively. Compare with the previous work, our experimental data is much more abundant, and the recognition rate is higher as well. Furthermore, with the proposed method, the reader could obtain fatty liver classifier easily.