The invention relates to a three-dimensional model simplification method based on appearance retention. The method comprises the following steps of: 1, reading model data; 2, dividing model top points into boundary points and non boundary points; 3, calculating a second order error metric matrix Q and other parameters of each top point; 4, calculating a folding value of each edge (vi, vj), and adjusting the folding value during calculation of the folding value according to boundary points, concave and convex changes and a detail image; 5, taking an edge with minimum folding value from a priority queue for simplification; and 6, judging whether a simplification requirement is met, if so, outputting a simplified model, otherwise, repeatedly executing a model simplification process from the step 2. The method retains high efficiency of second order error metric and model appearance consistency to the greatest extent, and is simple, practical, high in simplification speed and good in robustness.