Recent research highlights the important role of distinct emotions in politics. Yet, a majority of available dictionaries measure only negative versus positive tone in political text. Further more, dictionaries rely on the bag-of-words approach and are often tailored to specific domains and languages. This can lead to distorted results when applied to other contexts. This paper sets out to create, validate, and compare a novel emotional dictionary for the German political context. It can measure affective language in political communication associated with eight different emotions. Furthermore, to overcome limitations of the bag-of-words approach, the study further creates locally-trained word embedding models and compares their accuracy with the dictionary approach. The different approaches are validated on 10,000 crowd-coded train ing sentences. The results highlight important differences between the bag-of-words approach and word embeddings. Furthermore, both approaches deliver more accurate and robust results than available off-the-shelf dictionaries in measuring emotional language in German political communication.