The assessment and characterization of the endothelial function is a current research topic as it may play an important role in the diagnosis of cardiovascular diseases. Flow mediated dilatation may be used to investigate endothelial function, and B-mode ultrasonography is a cheap and non-invasive way to assess the vasodilation response. Computerized analysis techniques are very desirable to give higher accuracy and objectivity to the measurements. A new method is presented that solves some limitations of existing methods, which in general depend on accurate edge detection of the arterial wall. This method is based on a global image analysis strategy. The arterial vasodilation between two frames is modeled by a superposition of a rigid motion model and a stretching perpendicular to the artery. Both transformation models are recovered using an image registration algorithm based on normalized mutual information and a multi-resolution search framework. Temporal continuity of in the variation of the registration parameters is enforced with a Kalman filter, since the dilation process is known to be a gradual and continuous physiological phenomenon. The proposed method presents a negligible bias when compared with manual assessment. It also eliminates artifacts introduced by patient and probe motion, thus improving the accuracy of the measurements. Finally, it is also robust to typical problems of ultrasound, like speckle noise and poor image quality.