Social　bots　are　intelligent　programs　that　have　the　ability　to　receive　instructions　and　mimic　real　users’　behaviors　on　social　networks,　which　threaten　social　network　users’　information　security.　Current　researches　focus　on　modeling　classifiers　from　features　of　user　profile　and　behaviors　that　could　not　effectively　detect　burgeoning　social　bots.　This　paper　proposed　to　detect　social　bots　on　Twitter　based　on　tweets　similarity　which　including　content　similarity,　tweet　length　similarity,　punctuation　usage　similarity　and　stop　words　similarity.　In　addition,　the　LSA　(Latent　semantic　analysis)　model　is　adopted　to　calculate　similarity　degree　of　content.　The　results　show　that　tweets　similarity　has　significant　effect　on　social　bot　detection　and　the　proposed　method　can　reach　98.09%　precision　rate　on　new　data　set,　which　outperforms　Madhuri　Dewangan’s　method.　©　ICST　Institute　for　Computer　Sciences,　Social　Informatics　and　Telecommunications　Engineering　2018.