Direct　application　of　the　Kalman　filter　algorithm　to　state　estimation　of　civil　structures　presents　a　computational　challenge　due　to　their　high　dimensions　and　complexity.　Rewriting　dynamic　equations　using　modal　coordinates　can　be　an　alternative　solution　to　this　problem　because　high　modes　with　minimal　contributions　to　structural　responses　can　be　truncated.　Although　the　mode　selection　is　important　in　accurately　estimating　the　state　of　civil　structures,　studies　on　determining　the　remaining　mode　number　are　limited.　Hence,　the　mode　selection　method　in　the　Kalman　filter　for　the　optimal　reconstruction　of　structural　responses　is　investigated　in　this　study.　A　modal　signal-to-noise　ratio　(MSNR)　is　defined　as　the　ratio　of　the　estimated　modal　response　variance　to　the　corresponding　estimation　error　variance.　Only　modes　with　MSNR　values　higher　than　an　analytically　derived　threshold　are　selected.　A　beam　structure　is　numerically　investigated　to　examine　effects　of　excitation　amplitude　and　frequency,　measurement　noise,　and　number　of　sensors　on　the　adaptive　mode　selection　for　optimal　response　reconstruction.　Experimental　studies　using　a　simply　supported　overhanging　beam　also　confirm　the　efficacy　of　the　proposed　approach　in　response　reconstruction　using　multiple　types　of　sensors　(including　strain　gauges,　displacement　sensors　and　accelerometers).　Both　the　numerical　and　experimental　results　reveal　that　using　all　vibration　modes　or　a　complete　numerical　model　with　all　degrees　of　freedom　will　reduce　the　accuracy　of　response　reconstruction.　©　2021　Elsevier　Ltd