Auto-generation　of　Electronic　Health　Record　(EHR)　is　a　difficult　problem　in　intelligent　medical　diagnose　and　health　care.　This　paper　proposes　a　BiLSTM-CNN　attention　model　which　directly　reads　patients＇　complaints　and　generates　EHRs.　The　BiLSTM-CNN　attention　model　is　a　combination　of　BiLSTM　and　CNN　model　with　attention.　The　attention　is　achieved　through　the　Encode-Decode　model.　With　the　coded　input　text　the　BiLSTM-CNN　model　is　trained　and　used　to　generate　EHRs.　The　model　is　validated　against　reference　EHRs　which　shows　satisfactory　result.　The　ROUGE　is　also　used　as　the　evaluation　metrics　to　compare　with　other　baseline　models.　A　brief　discussion　about　the　limitations,　weakness　and　the　future　work　of　the　proposed　mode　are　given.　©　2019　IEEE.