Ultrasonic testing of austenitic stainless steel multipass welds is complex. Because the welded structure is both anisotropic and heterogeneous, the propagation of the ultrasonic beam is disturbed (attenuation, deviation, splitting), making diagnosis difficult. This diagnosis can be improved by modelling the ultrasonic propagation in the inspected weld. For this purpose, a description of the macrostructure is required. This can be obtained by optical means which provide a macrograph, but this involves a destructive cutting of the weld, or a weld sample. In the latter case, the sample must have been made and stored, and it must be representative of the inspected weld. As an alternative, it is possible to predict the macrostructure of the weld by using a numerical model more or less realistic. For example, Ogilvy’s model is an analytical model which predicts the macrostructure by considering a symmetrical structure of the weld. But this assumption is often non-realistic. Since 2000, EDF and the LMA develop another model called MINA which is a phenomenological model and so realistic one. MINA model predicts the macrostructure of Shielded Metal Arc Welding (SMAW) multipass welds, taking account information from welding conditions. It is then coupled to an ultrasonic simulation code, to simulate the impact of welding on ultrasonic propagation and therefore control. In our study the welds are made with Gas Tungsten Arc Welding (GTAW) process. In this paper we first show that existing models are not adapted to this welding process. Then, we present the adopted scientific approach and the first advances aiming at the development of a new model relevant for GTAW welds.