In　this　paper,　we　propose　a　new　method　based　on　comb-type　pilot　which　combines　the　Gaussian　process　regression　(GPR)　and　the　wavelet-denoising　for　channel　estimation　of　orthogonal　frequency-division　multiplexing　(OFDM)　system.　We　use　the　OFDM　receiver　to　estimate　channel　state　information　(CSI)　at　pilot　explicitly　and　detect/recover　the　transmitted　symbols　using　the　estimated　CSI.　The　initial　channel　frequency　response　on　the　pilot　position　is　obtained　by　Least　Square　(LS)　algorithm　and　processed　by　the　wavelet-denoising　algorithm　which　can　reduce　the　influence　of　noise　in　the　OFDM　communication　system.　Then　the　frequency　at　pilot　and　values　of　its　frequency　response　are　used　a　training　data　for　the　GPR　algorithm　to　get　the　channel　frequency　response　of　whole　carrier.　The　simulation　results　show　that　under　the　multi-path　channel　circumstance,　the　new　approach　significantly　reduces　the　bit　error　rates　(BER)　compared　to　traditional　method.　©　2019　IEEE.