Seismic source inversion is the method of inferring the spatial orientation of an earthquake source from seismic records. The results come with large uncertainties, which we try to estimate in a Bayesian approach. We propose an empirical relationship for the likelihood function based on a large dataset of deterministic solutions. This allows using the cross-correlation coefficient as a misfit criterion, which is better suited for waveform comparison than the popular root mean square or L2 norm.