Five　typical　partial　discharge　(PD)　models　of　oil-paper　insulation　are　made　in　the　article,　time　series　of　PD＇s　amplitude　are　fetched　from　the　results　measured,　after　being　pre-processed,　they　are　fitted　by　auto　regressive　(AR)　model,　the　coefficients　of　the　AR　model　are　used　as　feature　vector　to　recognize　the　five　typical　partial　discharges　by　BP　Neural　Network.　In　this　article,　different　order　AR　models　are　used　to　fit　the　time　series,　the　result　shows　that　under　the　4-order　or　6-order　AR　model,　the　recognition　accuracy　rates　of　the　five　models　are　all　as　high　as　over　80%.