This paper concerns contour-based algorithms for generating a 3D CAD model from medical images. The 3D model generated by contour-based algorithms can be used to generate CAM data for fabrication where the accuracy is of most concern. The overall procedure includes: (1) contour data extraction from medical images, (2) smoothing of the extracted contours, and (3) creation of a surface model from contours. For this, various methods should be applied to generate a highquality surface model. The main contribution of this paper is to propose a new contour smoothing method, called bi-directional smoothing. The basic idea behind the proposed bi-directional smoothing method is to refine contours along both (u, v) parametric directions. Compared to conventional smoothing methods, the recontouring that comprises the first part of the method can prevent the shape of a contour from shrinking with a large number of iterations. Along with recontouring, a vertical connectivity estimation and a vertical smoothing method are also proposed. The overall procedure for this approach is demonstrated with an application example using CT images of a femur.