Today, information obtained from such medical images as X rays, CT, and MRI, as well as ultrasonic images, is important for diagnosis. In particular, an ultrasonic image has the merits of safety, low cost, and simplicity of the machine. On the other hand, ultrasonic images are sometimes poor in quality because of electronic characteristics and speckle noise of the ultrasound equipment. Therefore, diagnosis by ultrasonic imaging has been dependent mainly on the judgment of doctors. It is necessary to develop an automatic diagnosis method based on qualitative evaluation. In this paper, we present a novel method of region extraction using statistical characteristics of the echo speckle pattern in ultrasonic imaging. The discriminating function is defined to recognize the similarity of the statistical characteristics. It is based on a multidimensional autoregressive model. Using the proposed region extraction method, it is possible to discriminate and extract different regions in one image. We also propose a method to extract the border of the tissue region as an application of the regional extraction method. Finally, we have evaluated these methods by analyzing ultrasonic images including regions of the tissue in the human body and simulated data.