The　dual-band　predistortion　models　suffer　from　high　complexity　and　non-adaptability　of　optimization　algorithms.　To　address　this　issue,　this　paper　proposes　an　adaptive　optimization　algorithm　for　dual-band　predistortion　model　with　reduced　complexity.　We　use　dual-band　general　memory　polynomial　(DB-GMP)　as　the　predistortion　model　where　all　basis　function　terms　of　the　original　DB-GMP　model　are　sorted　by　orthogonal　matching　pursuit　algorithm.　In　each　iteration,　all　selected　basis　function　terms　help　to　construct　an　alternative　model.　We　then　derive　the　Bayesian　information　criterion　(BIC)　when　output　vector　elements　of　the　DB-GMP　model　are　with　non-independent　identical　distributions,　and　the　model　with　smallest　BIC　value　is　treated　as　the　optimized　model.　Finally,　we　achieve　the　proposed　algorithm　without　the　information　of　model　sparsity　and　fitting　error　threshold.　Simulation　results　show　that　compared　with　the　original　DB-GMP　model,　the　coefficient　number　of　the　optimized　model　is　reduced　by　more　than　75%,　while　both　the　models　after　predistortion　have　almost　the　same　level　of　adjacent　channel　power　ratio　and　normalized　mean　squared　error,　leading　to　good　predistortion　performance.　©　2018,　Chinese　Institute　of　Electronics.　All　right　reserved.