In　recent　years,　several　substructural　identification　methods　have　been　developed　for　structural　health　monitoring.　Most　of　these　methods　are　deterministic,　and　unknown　parameters　in　the　target　can　be　identified.　However,　the　uncertainties　in　the　identified　results　cannot　be　evaluated.　This　paper　presents　a　Bayesian　probabilistic　model　updating　approach　for　substructure　identifi-cation.　A　new　response　reconstruction　technique　is　explored　and　combined　with　the　Bayesian　inference　method　for　probabilistic　model　updating　of　the　target　substructure.　The　large-scale　structure　was　divided　into　substructures,　and　the　uncertainties　in　the　identified　results　were　evaluated.　The　stochastic　gradient　descent　method　is　proposed　for　estimating　the　maximum　likelihood　estimation　and　maximum　a　posteriori　of　the　unknown　parameters　in　the　target　sub-structure.　The　posterior　distributions　of　the　unknown　parameters　are　estimated　using　an　asymptotic　approximation.　Numerical　studies　on　a　three-span　beam　structure　and　experimental　studies　on　an　eight-floor　steel　frame　were　conducted　to　verify　the　accuracy　and　efficiency　of　the　proposed　method.　The　results　show　that　the　estimated　results　match　the　actual　values,　and　reasonable　standard　deviations　can　be　obtained.