Every day a large amount of text is produced during public discourse. Some of this text is produced by actors whose political colour is very obvious. However, though many actors cannot clearly be associated with a political party, their statements may be biased towards a specific party. Identifying such biases is crucial for political research as well as for media consumers, especially when analysing the influence of the media on political discourse and vice versa. In this study, we investigate the extent to which political party affiliation can be predicted from textual content. Results indicate that automated classification of political affiliation is possible with an accuracy better than chance, even across different text domains. We propose methods to better interpret these results, and find that features not related to political policies, such as speech sentiment, can be discriminative and thus exploited by text analysis models.