Several methods for finite element modelling and analysis of tubular welded joints are described in various design codes. These codes provide specific recommendations for modelling of the welded joints, using simple weld geometries. In this paper, experimental strain range results from two full-scale semi-automatically welded K-nodes tests are compared to corresponding finite element models. Three dimensional (3D) scans of the weld surfaces have been developed for the automatically welded K-joints. These 3D scans are included in the finite element (FE) models to determine the accuracy of the FE models. Some discrepancies are observed between finite element predictions and test results. Finite element model updating has been applied to improve the accuracy of the developed finite element models. The results are compared to a FE model with a simple weld geometry based on common offshore design codes and a FE model without any modelled weld. The results show that the FE model with 3D scanned welds are more accurate than the other FE models. The precise location of weld toe in the 3D scanned weld is difficult to locate in the FE model, thus stochastic finite element modelling is performed to analyse the probabilistic hot-spot stresses. The results show large variations in the hot-spot stresses. It can be concluded that it is necessary to evaluate the probabilistic nature of stresses for calculating fatigue lives of welded tubular joints in offshore structures.