According　to　the　problem　of　fuzzy　rule　automatic　generation,　a　self-organization　stochastic　fuzzy　control　strategy　is　designed　by　adopting　OCPFA　learning　system　which　is　constructed　by　Skinner　operant　conditioning　(OC)　and　probabilistic　finite　automata　(PFA).　The　designed　OC　learning　mechanism　is　firstly　adopted　in　the　strategy　to　stochasticly　select　a　fuzzy　action　which　is　used　as　the　consequent　part　of　fuzzy　rule　from　the　fuzzy　action　sets;　then　the　OC　learning　mechanism　is　updated　by　using　feedback　information　of　selected　fuzzy　action　which　comes　from　environment;　finally　a　new　fuzzy　consequent　action　is　selected　based　on　the　updated　OC　learning　mechanism.　The　process　of　self-learning　and　self-organizing　are　theoretically　proved　convergent　in　sense　of　probability.　Both　the　simulation　and　experiment　indicate　that　the　stochastic　fuzzy　control　strategy　can　be　successfully　applied　in　self-balancing　control　of　two-wheeled　robot　and　the　learning　rate　and　control　precision　of　robot　are　improved.