Since　sensitivity　analysis　provides　information　on　changes　of　the　system　dynamics　with　respect　to　changes　of　system　parameters,　it　has　been　long　recognized　as　an　important　tool　for　power　system　small-signal　stability　analysis　and　controller　designs.　Sensitivity　analysis　is　now　much　more　important　due　to　all　the　anticipated　changes　of　future　power　systems,　so　more　efficient　and　feasible　methods　are　needed.　The　current　sensitivity　analysis　methods　are　either　simulation　based　or　eigenvalue　based.　The　eigenvalues　and/or　the　time　domain　trajectories　of　the　investigated　system　are　solved　for　scenarios　both　before　and　after　the　parameter　changes　to　evaluate　the　impacts　of　these　changes　on　the　system　dynamics.　However,　both　methods　are　actually　numerically　based,　so　often　insufficient　in　fully　revealing　system　physical　characteristics.　This　paper　analyzes　the　drawbacks　of　the　available　methods　for　analyzing　the　dynamic　impacts　of　parameter　changes　in　a　power　system　and　then　develops　a　feedback-based　model　for　sensitivity　analysis.　The　model　is　easier　to　identify　the　dynamic　interactions　of　components;　meanwhile,　it　also　offers　a　new　perspective　on　sensitivity　analysis　from　the　frequency　domain.　Case　studies　of　increasing　complexity　demonstrate　fully　that　the　application　of　the　proposed　method　reveals　new　insights　on　power　system　dynamics　that　conventional　methods　could　not　do.