The　quantitative　evaluation　of　the　importance　degree　of　spare　parts　is　essential　as　spare　parts＇　maintenance　is　critical　for　inventory　management.　Most　of　the　methods　used　in　previous　research　are　subjective.　For　this　reason,　an　accurate　method　for　the　evaluation　of　the　importance　degree　combining　an　improved　clustering　algorithm　with　a　back-propagation　neural　network　(BPNN)　is　proposed　in　the　present　paper.　First,　we　classified　the　spare　parts　by　analyzing　their　historical　maintenance　and　inventory　data.　Second,　we　evaluated　the　effectiveness　of　classification　using　the　Davies-Bouldin　index　and　the　Calinski-Harabasz　indicator　and　verified　it　using　the　training　data.　Finally,　we　used　BPNN　to　determine　the　training　data　necessary　for　an　accurate　assessment　of　the　importance　degree　of　spare　parts.　The　previous　importance　evaluation　methods　were　susceptible　to　subjective　factors　during　the　evaluation　process.　The　model　established　in　this　paper　used　the　actual　data　of　the　company　for　machine　learning　and　used　the　improved　clustering　algorithm　to　implement　training　and　classification　of　spare　parts　data.　The　importance　value　of　each　spare　part　was　output,　which　additionally　reduced　the　impact　of　subjective　factors　on　the　importance　evaluation.　At　the　same　time,　the　use　of　less　data　to　evaluate　the　importance　of　spare　parts　was　achieved,　which　improved　the　evaluation　efficiency.　©　2020　Shoujing　Zhang　et　al.