Direct empirical testing of forensic DNA micro-probabilities has proved difficult. Substantial support for the population genetic models on which these micro-probabilities are based has come from powerful but indirect means. However for many people empirical testing rather than models remains the ideal. Following the example set by Weir and Evett we report here a large-scale empirical test of a population genetic model incorporating substructure and relatedness. Given that all models represent simplifications of reality the fit of observed and prediction is remarkable and gives substantial support to the reliability of current methods.