We　propose　an　efficient　scheme　for　compressing　mesh　sequence　with　different　connectivities　using　a　thin-plate　spline　(TPS)　based　multi-reference　prediction　coding.　The　scheme　combines　the　single-reference　prediction　with　the　P-structure　prediction,　and　predicts　each　frame　of　the　sequence　using　the　TPS　transformation.　Compared　with　traditional　compression　scheme,　our　scheme　imposes　the　TPS　transformation　in　the　prediction　coding　which　utilizes　both　spatial　and　temporal　correlations　of　the　mesh　sequence　with　different　connectivities.　We　test　our　scheme　for　3D　facial　meshes　and　experimental　results　show　that　our　scheme　produces　both　accurate　reconstruction　meshes　and　high　compression　efficiency.　©　2016　NSP.