As　conventional　fault　recognition　methods　have　some　difficulties　such　as　low　accuracy,　time-consuming,　and　non-quantified　results,　we　propose　a　new　method　based　on　affinity　propagation　clustering　and　principal　component　analysis　(PCA)　in　this　paper.　At　first,　discontinuous　points　of　horizons　are　located　by　the　connected　component　labeling　method.　Then,　the　discontinuous　points　are　clustered　by　the　affinity　propagation　clustering　algorithm　and　a　fault　is　determined　by　a　type　of　discontinuous　point　cluster.　In　this　way,　both　the　number　of　faults　and　the　centers　of　all　the　cluster　types　are　obtained.　At　last,　the　principal　directions　of　the　discontinuous　points　of　all　the　cluster　types　are　determined　by　the　PCA.　A　direct　line　along　the　principal　direction　which　passes　through　a　cluster　center　is　considered　as　a　fault.　The　proposed　algorithm　and　conventional　methods　are　tested　on　model　and　real　seismic　data,　and　the　rationality　of　the　proposed　method　is　validated　by　the　peak　signal-to-noise　ratio,　mean　square　error,　time　consumption,　and　coincidence　rate　for　the　number　of　faults.　In　addition,　fault　quantitative　interpretation　achieved　by　the　proposed　algorithm　possesses　great　practical　significance　in　the　seismic　exploration.　©　2017,　Editorial　Department　OIL　GEOPHYSICAL　PROSPECTING.　All　right　reserved.