The　naissance　of　cognitive　robotics　marks　that　psychology　is　more　and　more　highly　involved　in　the　artificial　intelligence　research.　Inspired　by　psychology　and　ethology,　we　propose　an　operant　conditioning　learning　model　based　on　BP　(back-propagation)　network　named　OCLMBP　on　the　basis　of　Skinner＇s　relevant　theory.　The　model　is　applied　to　the　problem　of　obstacle　avoidance　with　a　wheeled　robot.　The　robot　controlled　by　the　model　can　learn　to　avoid　obstacles　through　a　learning-by-doing　style　without　any　external　supervision,　but　by　the　proximity　sensors　information　as　positive　or　negative　reinforcement　signals.　The　results　are　compared　with　original　OCLM　(operant　conditioning　learning　model),　and　the　proposed　model　has　better　performance.