Focusing on the maintenance of important shape features during the mannequin simplification process, a new simplification method, driven by fuzzy partition of geometric information gain metric is proposed in this paper. The integration of distance, non-planarity, change of the normal has been defined as the geometric information gain metric to measure the geometric information content of each vertex. The geometric information gain metric is subsequently partitioned into three fuzzy sets to indicate the weak, moderate and strong geometry salience of vertices on the underlying mannequin in which thresholding values for fuzzy sets are selected by minimum fuzziness degree principle. The resulted fuzzy sets of geometric information gain metric are then introduced to the edge collapse operation to determine the vertices incident to candidate edges for contraction. The proposed approach is capable of reducing simplification error and maintaining mannequin’s feature dimensions as well. Simulation results show that the new algorithm facilitate better approximations with regard to both visual fidelity and geometric errors than Garland’s quadric error metric algorithm.