Under　the　cloud　mode,　it　is　difficult　to　select　an　appropriate　shaft　part　processing　service　from　the　rich　and　huge　cloud　manufacturing　services.　The　K-means　algorithm　and　the　extension　theory　were　applied　into　the　cloud　platform　to　solve　this　problem.　The　similarity-based　clustering　method　was　proposed　to　preprocess　the　shaft　part　processing　cloud　services　and　generate　a　plurality　of　class　clusters.　Based　on　the　extension　theory,　the　matter-element　models　of　service　requirement　and　class　clusters　of　shaft　part　processing　cloud　services　were　established,　and　the　selection　method　of　shaft　part　processing　cloud　service　sets　was　presented.　The　result　of　the　study　shows　that　the　clustering　and　selection　method　of　shaft　part　manufacturing　services　on　cloud　platform　is　feasible　and　effective.