Since a few decades volcanic Long Period (LP) events have been recorded on many active volcanoes and their study has been recognized as an important tool to characterize volcanic activity. LP event analyses through moment tensor (MT) inversions have led to kinematic descriptions of various source mechanisms. The main challenge in these inversions is to " strip out " the propagation effect in order to isolate the source; hence the velocity model used controls the accuracy of the retrieved source mechanism. We first carry out several synthetic tests of inversions on Mt. Etna volcano (Italy). Four geological models with topography are considered with increasing complexity: the most complex model is used to generate synthetic data, while the other three models are used to calculate the Greens' functions for inversions. The retrieved solutions from the three velocity models are similar. The MT solutions for a deeper source are well retrieved, while a shallower source test suffers from high uncertainties and strong misinterpretation of the source orientation. The homogeneous model gives the lowest misfit value, but source location and mechanism decomposition are inaccurate. When a complex model different from the true one is used, a high misfit value and a wrong solution is obtained. We then incorporate our findings into the MT inversion of an LP event recorded on Mt Etna in 2008. We obtain very different solutions among the three models in terms of source location and mechanism decomposition. The overall shape of the retrieved source time functions are similar, but some amplitude differences arise, especially for the homogeneous model. Our work highlights the importance of including the unconsolidated surface materials in the computation of Green’s functions especially when dealing with shallow sources.