A mock-up of an offshore structure was prepared by multi-pass welding of several components with different thicknesses, different materials, different grooves, and ultra-long welding lengths. It may be very time consuming to obtain the stress distribution of the mock-up with conventional thermal elastic-plastic (TEP) computational methods. An efficient computation method, i.e. the model separation and stress assembly method, was proposed in the present study to obtain the stress distribution of the mock-up within an acceptable time. The full finite element (FE) model with solid elements was first created and separated into two independent parts, and the stress distribution in each part was obtained by using the TEP FE method. Finally, the full stress distribution in the mock-up was obtained by assembling the stress distributions from each part. The computed results show that the predicted stresses of the mock-up agree with the measured data obtained by using the hole-drilling method and x-ray diffraction method. Therefore, the proposed efficient method for stress simulation in large and complex structures can guarantee the simulation accuracy within an acceptable computation time on a common computer workstation. 1. Introduction Because of the intense concentration of heating during fusion welding, the welding seam and its vicinity undergo rapid heating and cooling, generating residual stress in the joint. Welding residual stress can be detrimental to the structure's performance because of fatigue, creep, and plastic collapse (Withers 2007). In addition, it can induce stress corrosion cracking (Dong et al. 1997). Therefore, investigation of welding residual stress distribution is very important to facilitate the structure design and life evaluation of welded structures. The experimental measurement of residual stress has practical limitations. For large and complex structures such as offshore components, it is impossible to obtain the full residual stress distribution from experiments. The finite element (FE) numerical simulation of the welding process can measure the full stress distribution during the welding process with the advantages of being economical, nondestructive, and repeatabile. Therefore, it has been widely applied in many industrial fields to investigate the mechanisms of welding processes, stress and distortion characteristics, and the service life of welded structures (Lindgren 2006).