Abstract: Welding-induced initial imperfections significantly influence the ultimate strength of hull structures. Conventional ultimate strength analyses often simplify or neglect these imperfections, introducing a considerable uncertainty. One thermal elastic-plastic finite element method is employed to simulate fillet welding in stiffened panels, and to obtain the three-dimensional welding distortion and residual stress as initial imperfections for the ultimate strength analysis. And experiments are conducted for a validation. Various imperfection evaluation methods (including the buckling mode method, empirical formulas and thermal elastic-plastic finite element method) are compared for their characteristics and impact on ultimate strength analyses. The effects of imperfection types and of loading conditions on the defective stiffened panel’s ultimate strength are also investigated. Research results show that the imperfection evaluation methods considerably influence ultimate strength analyses. The thermal elastic-plastic finite element method accurately captures welding-induced imperfections, showing a good agreement with experimental results. The welding boundary conditions require a careful consideration. For the model studied, the residual stress exhibits a more pronounced effect on ultimate strength than that of initial distortion, but their coupling effect does not have a more adverse impact. Under varying loading conditions, changes in the transverse residual stress distribution are crucial to ultimate strength and to stress-strain curves.