A time-honored maxim says that the judicial system is the last line of defending justice. Its performance has a great impact on how the citizen trust or distrust their state apparatus in a democracy. Technically speaking, the judicial process and its procedures are very complicated and the purpose of the whole system is to go through the law and due process to protect civil liberties and rights and to defend the public good of the nation. Therefore, it is worthwhile to assess the performance of judicial institutions in order to advance the efficiency and quality of judicial verdict. This paper combines data envelopment analysis (DEA) and decision trees to achieve this objective. In particular, DEA is first of all used to evaluate the relative efficiency of 18 district courts in Taiwan. Then, the efficiency scores and the overall efficiency of each decision making units are then used to train a decision tree model. Specifically, C5.0, CART, and CHAID decision trees are constructed for comparisons. The decision rules in the best decision tree model can be used to distinguish between efficient units and inefficient units and allow us to understand important factors affecting the efficiency of judicial institutions. The experimental result shows that C5.0 performs the best for predicting (in) efficient judicial institutions, which provides 80.37% average accuracy.