Abstract: For a robot remote welding system working in an unstructured environment,the coherence between virtual environment(VE) and real environment determines the reliability,operation accuracy and work efficiency of the system directly.The VE calibration technique can calibrate the relationship between robot model and working environment model in VE by the data from the real environment of remote worksite.The task of VE calibration in remote welding system is achieved by camera calibration and part localization which both rely on image feature acquisition in human-machine interaction mode and nonlinear least-squares algorithms combined with linear ones.In the experiment,the maximum error is 2.9 mm and the average error is 1.5 mm when a rather large focal length is used.An analysis of factors influencing the calibration accuracy is done based on the experimental process and results.