P300　brain-computer　interface　(BCI)　is　an　important　field　of　brain　science　exploration,　but　the　calibration　of　P300　affects　its　application.　To　solve　this　problem,　we　propose　an　algorithm　that　combines　transfer　learning　and　reinforcement　learning.　In　the　reinforcement　learning　algorithm,　we　refer　to　P300　linear　upper　confidence　bound(PLUCB).　Due　to　the　particularity　of　the　PLUCB　algorithm,　we　modify　it　and　integrate　the　idea　of　online　transfer　learning.　The　new　algorithm　is　applied　to　the　calibration-free　classification　of　P300　BCI,　using　the　classifier　matrices　of　the　subjects　in　the　source　domain,　without　collecting　additional　session　data　of　the　target　subjects　for　calibration.　We　test　the　performance　of　the　classifier　at　different　stages　of　the　algorithm.　For　each　subject,　the　agent　constantly　updates　on　the　first　part　of　the　data　and　the　second　part　of　the　data　is　used　for　testing.　The　results　show　that　our　designed　algorithm　P300　Homogeneous　Online　Transfer　Learning　(PHomOTL)　has　better　performance　than　PLUCB,　transfer　PLUCB　(TPLUCB)　and　Stepwise　Linear　Discriminant　Analysis　(SWLDA).　When　10000　trials　are　used　for　training　and　the　remaining　5120　trials　are　used　for　testing,　the　average　P300　classification　accuracy　of　PHomOTL　is　73.15%　and　the　average　character　classification　accuracy　of　PHomOTL　is　79.46%.　©　2022　IEEE.