Surface definition, a process of defining three dimensional surface from volume data, is essential in three dimensional volume data rendering. The traditional method applies a three dimensional gradient operator to the volume data to estimate the strength and orientation of surface present. Applying this method to ultrasound volume data does not produce satisfactory results due to noisy nature of the images and the sensitivity of certain signals to the direction of insonation. We propose a Bayesian approach to the surface definition problem of ultrasound images, and study this approach in two dimensions. We formulate the problem as the estimation of posterior means and standard deviations of Gibbs distributions for boundary believability and normal direction. A set of filters of directional derivatives of Gaussians are used to measure the edge strength and orientation at multiple scales. The likelihood function is based on the measurement at the smallest scale. The prior distribution reflects shape properties at multiple scales. It uses a pyramid algorithm for contour analysis where the lengths of contours are computed and contour gaps are closed at multiple scales. The outcome of the pyramid algorithm is the length and weight global attributes for each pixel. These attribute values are incorporated into the Gibbs prior using a data augmentation scheme. The design and implementation of such an approach are the subject of this paper.