Parametric approach and geometric method are the two most commonly used methods in industrial robot calibration. Screw axis measurement is a major step of the geometric method, which is believed can provide the actual geometric information of the robot. However, to obtain better results, the data in a large motion range and abundant measuring points are needed when conducting the screw axis measurement, while traditional measurement devices (such as laser trackers, coordinate measuring machines, etc.) are time-consuming. The laser tracker may even take longer to search the spherically mounted retro-reflector (SMR) due to the limitation of the incident angle, sometimes may even lose the SMR’s position. In this article, we try to use a multi-camera system which is well known for its high flexibility and fast testing speed to perform the screw axis measurement of an industrial robot. Besides, we choose to identify the MDH parameters instead of the DH parameters to overcome the singularity of the DH modal when encountering the approximately parallel adjacent axes. Also, the dual vector is used to simplify the calculation and the description of the axis. Based on the redundancy of data during measurement, an iterative method was developed to reduce the impact of noise. After the proposed approach is verified in simulation, an experiment was carried out on real industrial robots. We found the identified kinematic parameters are in good agreement with those obtained using laser tracker while the test speed of the multi-camera system is much faster than laser tracker and much less expensive.