OBJECTIVE: Quadric Error Metrics (QEM) algorithm can simplify complex 3D models. However, for its simple error metrics, QEM is not suitable to simplify medical model that contains many minute detail features. We need to develop a new algorithm because we are often interested in these features. METHODS: Based on QEM, We classify the vertex into seven classes. Different class has different cost of collapse. In our algorithm, collapse usually occurs with the lowest cost during simplification, thus minute feature can be preserved as possible. We also introduce the average of correlative triangles' area to estimate the volume change during the simplification. RESULTS: We test our algorithm on several complex medical models and find that our improved algorithm is effective and fast. CONCLUSION: To simplify a complex medical model, we want to preserve the minute feature during simplification. By classifying the vertexes, we develop an improved algorithm based on QEM. The testing results shows that our algorithm not only maintains the high efficiency of QEM algorithm, but also meets the high requirements of medical image processing on fidelity and mesh quality.