Abstract: In order to synthesize a realistic expression for people with their neutral face models,we propose a new parallel 3D facial expression synthesis method.We firstly build a realistic expression model by integrating the principal geodesic analysis(PGA) with radial basis function(RBF) method.Then a Poisson-based deformation method is used to synthesize realistic expressions.The proposed approach is very suitable for acceleration by GPU parallel programming.It can effectively synthesize realistic 3D facial expressions with high-resolution.The experimental result on the public BU-3DFED database demonstrates that the execution speed implemented using GPU is 13 times faster than using Dual-core CPU.It achieves near real-time synthesis performance.