Robot kinematic calibration is a process that reduces the differences between the nominal robot and real robot. The nominal robot has the notional geometry, based on its own design specification. Their geometry is simple and based on several assumptions such as parallelism or orthogonality of the axes of the joints. But the geometry of the real robot is a little different from that of nominal robot. This is because of manufacturing tolerance, mounting errors during robot link assembly and inaccuracy between the robot and the workcell where all objects for a given task are located. It is important to know the accurate values of the parameters of the real robot for the high precision task. In general, this is done by external 3D position or position/orientation sensor. From this information, we can find the kinematics of the real robot. Inverse kinematics, however, would not be calculated because the procedure is nearly impossible and tedious. To solve this problem, a compensation method is usually used. In this paper, we propose a new parameter identification algorithm for the robot kinematic calibration, which is very simple and applicable to off-line simulator and explain the remainder steps, modeling, measurement and compensation. Particularly, the pure data is extracted from the sensor data with noise by the method of MSE(Mean Square Estimation) and/or MLE (Maximum Likelihood Estimation) based on Bayesian rule.