For　traditional　FE　model　updating　methods　facing　numerous　unknown　parameters,　considerable　computation　efforts　are　always　required　and　worse,　an　ill-conditioned　optimization　problem　may　occur　without　convergence.　To　overcome　such　drawbacks,　one　may　consider　using　relatively　simple　mathematical　models　as　surrogates　for　FE　models　during　updating.　This　study　introduces　response　surface　models　to　model　updating　for　their　fast　computation　and　easy　implementation.　Such　models　can　correlate　the　inputs　of　a　physical　system　with　the　outputs　by　explicit　mathematical　expressions,　which　replace　the　original　FE　model　for　computation　during　updating.　By　this　means,　the　construction　of　sensitivity　matrices　is　avoided　and　the　computation　cost　considerably　decreases　with　high　convergence　speed.　However,　so　far　the　feasibility　and　cost-efficiency　of　response　surface　based　model　updating　has　not　been　fully　validated.　Therefore,　in　this　study,　the　analysis　of　variance　is　first　used　to　investigate　the　sensitivity　of　responses　to　updating　parameters.　Then　first-order　linear　models　are　constructed　using　the　D-optimal　design　requiring　a　nearly　minimum　number　of　samples.　The　Young＇s　moduli　and　modal　frequencies　are　adopted　as　the　model　inputs　and　outputs　respectively.　And　the　original　FE　models　are　then　replaced　by　the　response　surface　models　during　the　iteration　process.　Lastly,　the　proposed　method　has　been　validated　on　an　experimental　reinforced　concrete　frame　and　based　on　the　measured　data　of　the　damaged　frame,　the　damaged　regions　can　be　well　located　with　the　severities　identified.