The　increasing　incidence　of　depression　has　attracted　increased　attention　to　mental-health　document　retrieval　techniques　which　aims　to　help　individuals　efficiently　locate　documents　and　resources　relevant　to　their　depressive　problems.　However,　current　retrieval　systems　generally　have　low　accuracy.　We　propose　combining　a　Valence-Arousal-based　(VA-based)　retrieval　model　and　other　word-based　retrieval　models　to　improve　the　precision　of　retrieval　results.　The　VA-based　retrieval　model　considers　affective　words　extracted　from　queries,　which　help　provide　a　better　understanding　of　user　queries.　Experimental　results　demonstrate　that　the　combined　methods　outperform　the　word-based　retrieval　models　which　adopt　word-level　information　alone,　such　as　vector　space　model　and　BM25　model.