The performance of local tax offices of Norway is studied over a three-year period applying Data Envelopment Efficiency analysis and a Malmquist productivity index. The estimates are bias-corrected using a bootstrap approach recently developed for DEA models. The results show that bias correction and the construction of confidence intervals give a quite different picture without bootstrapping. A set of best practice offices is identified for future work on finding explanations for good performance. The productivity development of individual offices is classified into the four categories: productivity improving cost increase, productivity improving cost savings, productivity decreasing cost savings and productivity decreasing cost increase.