Facing the precision problem in flexible vision measurement caused by p ositioning error of robot,this paper studies a new method based on vision technique to compensate the industria l robot′s positioning error. Optical control points and two large field cameras are separately set on and beh ind the vision sensor to real-timely measure the position of the vision sensor,then to compensate the positioning er ror of robot.Dual-camera method′s precision is high,but it′s difficult to guarantee intervisibility in practica l application.A single orientation camera is attached to sensor,and global control points are set in the measured field of th e sensor.The global control points are measured by orientation camera to obtain the transformation relation between the current position of sensor and the global coordinate system,finally to compensate the positioning error.But the c ompensation precision is low in direct application.According to the source and features of error,this paper im proves the compensation method.By regarding angle parameter as a known quantity,we only optimize the displacement variables.Experiment results show that the root-mean-square error (RMSE) of single camera compensation meth od is 3.422mm,while that of improved method is less than 0.1mm.The improved compensation method enhances sensor′s positio ning pre cision.Because of its simple model and feasibility,it can satisfy the precision requirement for compensation of industrial robot.