In this paper, multiresolution models are employed in the context of reverse engineering for feature line extraction. Starting with a proper triangulation of the cloud point data as a priori, our feature line extraction algorithm has three steps: (1) establishing a Gauss normal sphere and creating multiresolution models for the Gauss sphere based on different levels of subdivisions for the sphere regions; (2) mapping the unit normal vectors of triangular faces in the multiresolution Gauss sphere and merging those connected triangular faces whose unit normal vectors fallen on the same Gauss sphere region with a given resolution to form super-faces (the collection of triangular facets with similar normal); and (3) extracting the boundaries from the super-faces and generating the feature lines from the extracted boundaries. We then use these feature lines as a base for tracing boundary curves for B-spline surface construction. Since feature lines maintain the characteristics of the original product models, in this way, we have a good chance to reconstruct B-spline surfaces with high quality. Examples are given in the paper to show the feature line extraction, the influence of the feature lines extracted under different resolutions, and the final reconstructed B-spline surfaces based on these feature lines.