Welding processes may cause undesirable residual stresses and their detection is possible using different methods. Some residual stress detection methods are destructive or semi-destructive, therefore not always applicable. Some methods are non-destructive, but complex and high cost. One possible non-destructive method to detect welding residual stresses, lower cost than the current techniques, is being studied by some researchers. This method is based in the phenomenon which a structure natural frequencies vary when welding residual stress is applied. It consists in measuring the structure natural frequencies after welding, and comparing them to natural frequency values considered ideal values obtained from finite element method simulation. However, the welding process simulation is not trivial; it implies at least four finite element analyses which can be executed in different manners. The aim of this study is to analyze which parameters can improve FE model performance for obtaining simulation results closer to experiments. In this study, five models were developed. It was observed that models using 2D shell elements generate better results than models using 3D solid elements. In addition, it was observed that the symmetry technique which can be used in plate simulations leads to significant lower computational times, but affect modal results and, in addition, do not generate natural frequency values for all vibrational modes; therefore, the symmetry technique should be avoided in this type of analysis. The birth and death technique, which simulates filler metal deposition, was also analyzed. Finally, this work proposes to use interpolation technique for the natural frequency values to evaluate the modal results variation due to welding residual stresses.