Twitter has become an important platform for political discussions among both politicians and the public and was extensively used during the 2021 federal election in Germany. Previous research examined the sentiment of the major political actors during that election on Twitter, but it remains unclear how the German public responded to them on Twitter in terms of sentiment. We analyzed a corpus of 713,742 tweets mentioning the Twitter handle of 89 of the most important party and politician accounts. We annotated a subset of 2,000 of these tweets regarding their sentiment and used this and other annotated corpora to implement and evaluate sentiment analysis algorithms based on singlelabel classification (positive, negative and neutral). We achieved best results with the German BERT model gbert-large using a combination of our annotated corpus and a previously annotated corpus from the same context as training material. This model achieves an average accuracy of 81.8% in a 5x5 cross-validation setting. Applying sentiment analysis on the overall corpus revealed that the majority of the tweets expressed negative sentiments. We investigated sentiment developments per party and show that sentiment was driven by significant events such as the implementation of stricter COVID19 regulations.