Numerous　software　reliability　growth　models　(SRGMs)　are　proposed　in　the　last　four　decades　to　track　the　reliability　growth　of　software　during　the　testing　phase.　In　many　existing　models,　it　is　assumed　that　the　faults　in　the　software　are　of　the　same　type.　However,　this　assumption　may　not　be　realistic.　Software　contains　different　faults　and　the　detection/correction　of　different　faults　can　require　different　efforts.　In　this　paper,　we　propose　a　more　general　software　reliability　model　considering　heterogeneous　faults,　where　the　parameters　for　the　detection　time　and　correction　delay　of　different　faults　are　assumed　to　be　faults　specific　and　allowed　to　follow　arbitrary　distributions.　Considering　different　parameters　makes　the　model　more　flexible　and　adaptable　to　different　practical　software　development　situations.　Some　specific　models　are　derived　under　different　assumptions　on　the　distributions　of　fault　detection　time,　fault　correction　delay,　and　the　corresponding　parameters.　Both　MLE　and　LSE　are　proposed　for　parameter　estimation.　Illustrative　examples　with　real　data　set　are　studied　to　compare　the　proposed　models　with　homogeneous　faults　model,　and　the　results　have　shown　advantages　of　our　models.　The　optimal　software　release　policy　based　on　the　proposed　model　is　also　studied.