The spatial slip distribution of the August 17, 1999, Izmit (Turkey) earthquake has previously been determined from both geodetic and seismological data. Though the models derived agree on a broad scale, they are significantly different in detail. These differences could be due to the inversion techniques applied and/or additional constraints implemented and/or be due to poor resolution. It is also conceivable that the relative agreement at larger scales results in part from a resolution artifact common to all models. To investigate this we perform a resolution analysis of fault slip models based on the inversion of surface displacement data. Assuming the theory of elastic dislocations in a half-space and adopting a parameterization for which model predictions are capable of fitting the spatial variations of the data, the inverse problem proves to be intrinsically ill-conditioned.