The　approximation　of　the　probability　density　function　(PDF)　of　the　structural　response　is　importance　in　structural　reliability　analysis.　The　recently　developed　cubic　normal　distribution　can　be　used　to　represent　the　PDF　of　structural　response　due　to　its　high　flexibility　and　its　large　applicable　range,　whose　parameters　are　evaluated　by　its　first　four　moments.　Therefore,　efficient　estimation　of　the　moments　is　of　great　importance.　Although　some　methods　have　been　developed,　evaluating　the　moments　of　structural　response　from　the　sight　of　balancing　accuracy　and　efficiency　remains　a　challenge,　especially　when　the　structural　response　is　implicit　and　high-dimensional.　In　this　paper,　based　on　the　artificial　neural　network　(ANN),　a　new　method　is　developed　to　efficiently　estimate　the　statistical　moments　of　structural　response.　The　main　procedure　of　the　proposed　method　includes　two　steps:　the　structural　response　is　approximated　by　the　ANN　and　then　the　moments　of　structural　response　can　be　easily　obtained.　A　RC　frame　structure　with　non-linear　behavior　is　used　to　demonstrate　the　efficiency,　accuracy,　and　applicability　of　the　proposed　method.　The　results　show　that　the　proposed　method　is　of　high　accuracy　and　efficiency　and　provides　a　robust　tool　for　representing　the　PDF　of　structural　response.　©　13th　International　Conference　on　Applications　of　Statistics　and　Probability　in　Civil　Engineering,　ICASP　2019.　All　rights　reserved.