Network　tomography,　a　combination　of　network　measurement　methodology　and　statistical　inference　strategy,　is　a　novel　research　field　of　network　measurement.　It　can　provide　much　more　accurate　information　about　link-level　performance　than　existing　techniques　widely　used　in　network　measurement.　This　paper　deals　with　the　problem　of　inferring　link　delay　characteristics　based　on　end-to-end　measurements　of　multicast　traffic.　Under　the　assumptions　that　link　delays　are　spatially　and　temporally　independent,　an　approach　is　proposed　to　discretize　link　delay,　including　subsection　discrete　case　and　uniform　discrete　case.　Then　it　is　estimated　link　delay　distributions　by　using　Maximum　Likelihood　Estimation　(MLE)　via　Expectation　Maximization　(EM)　in　terms　of　neighboring　link　delay　distributions.　The　results　of　OPNET　simulation　are　present　to　demonstrate　accuracy　of　link　delay　inference　procedure.　©　2008　Binary　Information　Press.