In recent years the economic performance of public non-profit sectors such as cultural services has become an interesting economic issue. This is due to the high dependence of cultural institutions on public funding on the one hand and the increasing cost-pressure on public budgets on the other hand. In order to achieve an efficient, cost-minimizing resource allocation public authorities who decide on the distribution of public budgets need reliable performance indicators. Against this background, this paper analyzes the efficiency of German public theaters for the seasons 1991/1992-2005/2006. Using a stochastic frontier analysis approach, we test whether the assumption of cost-minimizing behavior is reliable in this sector. Moreover, several panel data models that differ in their ability to account for unobserved heterogeneity are applied to evaluate the impact of unobserved heterogeneity on the efficiency estimates. The results indicate that the cost-minimizing assumption cannot be maintained. Consequently, an efficiency analysis based on a cost function approach seems inappropriate in the case of German public theaters. Further, we find a considerable unobserved heterogeneity across the theaters, which causes a significant variation in the models' efficiency estimates. This implies that failing to account for unobserved heterogeneity leads to biased efficiency values. Overall, our results suggest that there is still space for improvement in the employment of resources in the sector.