Predicting the outcome or the probability of winning a legal case has always been highly attractive in legal sciences and practice. Hardly any attempt has been made to predict the outcome of German cases, although prior court decisions become more and more important in various legal domains of Germany’s jurisdiction, e.g., tax law.This paper summarizes our research on training a machine learning classifier to determine likelihood ratios and thus predict the outcome of a restricted set of cases from Germany’s jurisdiction. Based on a data set of German tax law cases (44 285 documents from 1945 to 2016) we selected those cases which belong to an appeal decision (5 990 documents). We used the provided meta-data and natural language processing to extract 11 relevant features and trained a Naive Bayes classifier to predict whether an appeal is going to be successful or not.The evaluation (10-fold cross validation) on the data set has shown a performance regarding