This　paper　constructs　a　stochastic　learning　automaton　that　can　respond　the　operant　conditioning　behavior　based　on　probabilistic　automata,　which　is　used　for　simulating　skinner-pigeon　experiment.　The　stochastic　learning　automaton　is　a　kind　of　intelligent　unit　which　can　accomplish　adaptive　decision　under　unknown　environment,　and　so　it　can　let　an　agent　to　adapt　its　actions　to　gain　maximally　from　the　environment　while　only　being　rewarded　for　correct　performance.　A　stochastic　learning　automation　model　is　established　to　be　applied　to　skinner-pigeon　experiment　of　the　peck　button　task.　The　pigeon　learns　this　task　in　stages.　In　simulation,　the　model　also　acquires　the　task　in　a　similar　manner.　The　stochastic　learning　automaton　has　outstanding　superiority　in　dealing　with　the　problem　of　lack　of　prior　knowledge,　which　lays　a　theoretical　foundation　for　copying　the　behaviors　of　people　and　animal　by　robot　learning.