The　traditional　data　mining　is　unable　to　solve　the　problem　of　distributional　magnanimous　data　mining　and　the　distributional　system　is　very　difficult　to　solve　the　question　of　heterogeneous　operating　systems　and　proposals.　Under　the　grid　environment,　parallel　data　mining　can　solve　heterogeneous,　distributed　and　magnanimous　data　mining　problem.　In　this　paper　the　existing　strategy　and　the　association　rule　algorithm　of　parallel　data　mining　is　analyzed　and　a　new　association　rule　parallel　mining　plan　is　proposed　based　on　the　grid　environment.　The　new　strategy　is　based　on　the　agent　technology,　multi-thread　and　the　centralism　vote　technology.　The　experimental　results　in　a　simulated　grid　platform　show　the　feasibility　and　the　effectiveness　of　the　new　strategy.