This　letter　proposes　a　novel　electromagnetic　(EM)　optimization　approach　using　adjoint-sensitivity-based　multifeature　surrogate　for　microwave　filter　design.　The　proposed　technique　aims　to　solve　the　EM　optimization　problem,　in　which　the　starting　point　is　far　from　the　design　specifications.　We　propose　to　add　feature　sensitivities　in　multifeature　surrogate　modeling　to　achieve　a　more　accurate　surrogate　compared　with　the　existing　methods　without　feature　sensitivities.　We　also　propose　a　new　design　objective　function　in　feature　space　to　further　improve　the　efficiency　of　multifeature-assisted　optimization.　Two　microwave　filter　examples　are　used　to　demonstrate　the　proposed　technique.