A　domain　ontology　consists　of　two　major　components:　A　set　of　domain-specific　concepts　and　a　set　of　semantic-relations　between　these　concepts.　The　semantic-relations,　also　called　as　ontology　relations　are　used　to　support　semantic　retrieval　and　knowledge　management　in　ontology-based　applications.　To　reduce　difficulties　in　manual　ontology-building,　this　paper　proposes　a　semantic-relation　learning　method　for　the　purpose　of　automatically　discovering　relations　between　concepts.　Given　the　set　of　domain-specific　concepts　and　a　domain　corpus,　the　method　firstly　converts　concepts　into　feature-vectors　based　on　their　local　context　and　then　calculates　relevance　degrees　between　each　pair　of　concepts　based　on　the　similarity　of　their　feature-vectors　to　discover　related　concepts.　The　method　is　compared　with　current　state-of-the-art　in　two　ways:　(a)　differences　between　learning　results　and　the　golden　standard　defined　by　domain　experts,　and　(b)　differences　between　learning　results　and　the　standard　defined　by　the　CNCTST　(China　National　Committee　for　Terms　in　Sciences　and　Technologies).　Experiments　show　that　the　proposed　method　is　much　better　than　currently　existing　ones,　especially　in　term　of　recall　rate,　and　has　good　potentials　for　applications　such　as　ontology　building,　text　mining　and　semantic　retrieval.