Earthquake source inversions use data recorded at seismic stations to infer the details of the earthquake rupture process, while assuming that the Earth structure is known. The goodness of the source model is commonly assessed based on the goodness of the fit between recorded and synthetic ground motion. Given the sophisticated algorithms and advanced computer power available nowadays, finding source models that generate good fits to data is a rather easy task. However, different source models often fit data equally well. A more challenging task is to assess how good each source model is, i.e., to assess the uncertainty associated with each source model. In this talk I will present the problem of uncertainty in earthquake source inversions, both in the case of point-source moment tensor inversion and in the case of kinematic finite-fault inversions. I will then present strategies to assess uncertainty in source models. Finally, I will show how using different datasets in the inversion helps constrain the source models, thus reducing their associated uncertainty.