This　paper　proposes　a　new　algorithm,　called　TAM-WAP(the　short-hand　for　Top-down　Algorithm　for　Mining　Web　Access　Patterns),　to　mine　interesting　WAP　from　Web　logs.　TAM-WAP　searches　the　P-tree　database　in　the　top-down　manner　to　mine　WAP.　By　selectively　building　intermediate　data　according　to　the　features　of　current　area　to　be　mined,　it　can　avoid　stubbornly　building　intermediate　data　for　each　step　of　mining　process.　The　experiments　for　both　real　data　and　artificial　data　show　that　our　algorithm　outperforms　conventional　methods.　©　Springer-Verlag　Berlin　Heidelberg　2005.