The affine transformation of complex mesh model is implemented on the programmable GPU in order to speed up the calculation process. Based on the characteristics of parallel computing, each thread calculates the weight of the k-nearest neighbor and the transformed coordinate for one node, as well as multithreads run at the same time. By arranging the thread hierarchy and distributing the device memory reasonably, the parallel operation capability of GPU is brought into full play. And the experimental result shows GPU executes the affine transformation of massive mesh model with a significantly accelerated effect.