Electromagnetic　(EM)-based　yield　estimation　plays　an　important　role　in　microwave　design　due　to　the　presence　of　uncertainties　in　manufacturing　processes.　In　this　paper,　we　propose　a　novel　training　approach　with　adaptive　weighting　factors　to　increase　the　yield　estimation　accuracy　of　microwave　components　using　space　mapping　(SM)　surrogates.　In　this　approach,　an　adaptive　weighting　factor　is　set　up　for　each　frequency　point　of　interest　based　on　the　sensitivity　degree　of　the　EM　response　relative　to　the　design　specification.　A　novel　error　function　incorporating　the　adaptive　weighting　factors　is　proposed　specifically　for　EM-based　yield　estimation.　Using　the　proposed　error　function　to　train　the　SM　surrogate　enhances　the　model　accuracy　at　the　key　frequency　points　where　the　EM　response　is　sensitive　w.r.t.　to　statistical　variables　while　preserving　the　model　accuracy　at　other　ordinary　frequency　points　over　the　whole　frequency　range　of　interest.　Compared　with　the　existing　training　method,　the　proposed　approach　achieves　higher　yield　estimation　accuracy　especially　for　microwave　circuits　with　high　sensitivities.　The　proposed　approach　is　illustrated　by　a　microwave　filter　example.