The possibility of using tomography data as a constraint in variational data assimilation is explored by performing twin experiments. Realistic travel-time data are generated in a quasi-geostrophic model applied to the western Mediterranean Sea. After checking the robustness of the method with this dataset, a sensitivity study analyses the propagation of information by the model, and tests the effect of different parameters such as the starting point of the optimization or the length of the assimilation period. It appears that the variational method is well adapted to the nature of tomographic data that provide at the same time high time resolution and space integrals. The results clearly show the ability of this dataset to recover the temporal evolution of heat content in each layer and improve various components of the circulation described by the model. Tomography must then be considered as an important source of information that may contribute with altirnetry or local in situ measurement to the description of the ocean state and its evolution.