A　method　for　inferring　network　link-level　delay　distributions　based　on　discrete　delay　model　was　proposed　to　accurately　detect　networkinternal　delay　characteristics　without　the　cooperation　of　network　elements.　The　method　was　based　on　end-to-end　measurements　of　multicast　traffic.　According　to　different　network　path-level　delay　distributions,　the　bin　sizes　and　the　bin　intervals　in　the　discrete　delay　model　were　selected　by　using　different　binning　patterns.　Network　path-level　delay　distributions　were　analyzed　by　maximum　likelihood　estimation　and　expectation　maximization　algorithm　to　infer　network　link-level　delay　distributions.　It　not　only　can　rationally　initialize　delay　distributions,　but　also　can　reduce　computational　times　for　estimation.　Simulation　results　show　the　proposed　method　can　accurately　infer　network　link-level　delay　distributions.