A snake-based approach for the automatic detection of the intima-media thickness (IMT) of the far wall of the common carotid artery from sonographic images is proposed. The main problems to be faced are related, from one point, to the high level of speckle noise and, from the other, to the need of an accurate segmentation of the vessel structure for diagnostic purposes. In particular, the detection of the intima layer plays a fundamental role as it represents the initial contour from where the method should start. We propose an automatic segmentation approach which makes use of a first non linear filtering based on anisotropic diffusion followed by an iterative relaxation procedure. Once the intima layer has been detected our method tries to locate an optimal initial contour to detect the wall of the artery by minimizing of a modified energy functional.