Grid　technology　is　popular　in　science　and　business　computing　environment,　with　workflow　the　flexibility　and　applicability　of　the　grid　system　can　be　greatly　enhanced.　Cost-effective　scheduling　of　grid　workflow　applications　represented　by　DAG(Directed　Acyclic　Graph)is　an　essential　and　complex　problem.　In　general　such　problem　is　NP-hard.　In　this　paper,　a　novel　heuristics,　DSL(Deadline　Segment　Leveling)　is　proposed.　Considering　its　parallel　and　synchronization　properties,　the　grid　workflow　application　is　divided　into　segments,　and　each　segment　is　partitioned　into　groups　further　using　a　segment　leveling　method.　The　floating　time　is　prorated　to　each　group　to　enlarge　group　duration.　In　term　of　control　dependencies　between　tasks,　time　interval　for　each　task　is　determined.　Experimental　results　shows　DSL　can　achieve　considerable　performance　improvement　than　MCP(Minimal　Critical　Path)　and　other　leveling-based　heuristics.