In　electromagnetic　(EM)　optimization　for　microwave　circuits,　a　poorly　chosen　initial　point　can　easily　lead　to　falling　into　local　optima　that　fail　to　satisfy　design　specifications.　This　paper　presents　two　recent　EM　optimization　techniques　utilizing　features　and　feature　sensitivities　to　address　this　challenge.　The　first　technique　extracts　features　and　their　sensitivities　in　advance　to　build　a　multifeature　surrogate　model,　enhancing　optimization　efficiency.　The　second　technique　calculates　feature　and　their　sensitivities　by　constructing　the　neuro-transfer　function　(TF)　surrogate　model　to　accelerate　gradient-based　optimization.　Each　technique　is　illustrated　with　a　microwave　filter　example.　©　2024　IEEE.