Test　data　automatic　generation　is　the　most　significant　processing　section　in　the　software　test.　GA(Genetic　algorithm)　has　been　widely　used　for　automatic　generation　to　reduce　the　number　of　grade　and　improve　the　test　efficiency.　Currently　it＇s　inefficient　and　hardly　applied　to　practice　that　most　of　the　GAs　only　solve　single-path　or　consider　path　coverage　ratio　as　criteria.　In　this　paper,　we　compared　different　execute　path＇s　similarity　with　both　path　coverage　and　branch　coverage,　redesigned　the　fitness　function　and　the　operating　operator　to　improve　the　ability　of　the　GA＇s　coverage　and　research,　proposed　an　improved　GA　to　raise　the　ratio　of　path　coverage.　Experimental　results　show　that　our　proposed　GA　has　a　good　efficiency.