An　improved　approximate　Bayesian　computation　method　is　developed　by　incorporating　approximate　Bayesian　computation　(ABC),　Metropolis　Hastings　(MH)　sampling　and　stochastic　response　surface　(SRS).　The　improved　method　aims　to　fast　estimate　the　posterior　probability　distributions　of　random　parameters,　which　are　further　used　to　identify　structural　damage.　Firstly,　ABC　is　combined　with　the　MH　sampling　where　the　concepts　of　＇proposal　distribution＇,　＇acceptance　probability＇　and　＇dynamic　tolerance＇　are　introduced　to　evaluate　the　sampling　accuracy.　Meanwhile,　in　order　to　avoid　solving　the　likelihood　function　of　the　Bayesian　formula,　an　error　function　is　established　to　evaluate　the　similarity　between　simulated　and　measured　samples.　Secondarily,　SRS　is　used　to　express　the　relationship　between　the　random　parameters　and　responses　of　a　structure.　Then　the　statistical　features　of　the　responses　can　be　fast　calculated　even　facing　a　large　number　of　samples.　Lastly,　the　improved　ABC　method　is　applied　to　evaluate　the　parameter　posterior　probability　distributions　of　a　reinforced　concrete　beam.　Probabilistic　damage　indices　before　and　after　damage　are　also　constructed　and　then　compared　in　order　to　identify　the　damage　locations　and　severities　of　the　beam.　The　analysis　results　demonstrate　that　the　improved　ABC　method　can　effectively　reduce　the　solution　difficulty　for　estimating　parameter　posterior　probability　distributions.　Simultaneously,　the　solution　efficiency　of　a　Bayesian　parameter　identification　problem　can　be　improved.　Most　important　of　all,　multiple　damage　locations　can　be　successfully　identified　by　the　improved　ABC　method,　instead　of　the　traditional　Bayesian　method.　©　2019,　Nanjing　Univ.　of　Aeronautics　an　Astronautics.　All　right　reserved.