Information pertaining to real world problems often contains noises and uncertainties. This has been a major challenge faced by the contemporary AI researchers. Of various paradigms developed for handing uncertainties, the Dempster-Shafer theory (DS) and the Bayesian Belief Networks (BBN) have received considerable attention in the AI community recently. They have been successfully applied to problems in medical diagnosis, decision-making, image understanding, machine vision, etc.. Despite their obvious success, blindly using them without understanding their limitations may result in computational difficulty and unsatisfying inference results. The aim of this paper is to analyze and compare the performance of the two paradigms in extracting manufacturing features from the solid model descriptions of objects. Such a comparison will serve to identify their strengths, weakness, and appropriate application domains.