Automatically assigning POStags to languages for which little annotated training data is available is a challenging task in computational linguistics. When developing a POStagger for historical data one is confronted with an additional difficulty: a large variation in spelling. To tag historical Dutch texts, researchers often resort to taggers trained on modern Dutch data, although their adequacy for historical corpora is highly questionable. We present an analysis of this adequacy on 17th century Dutch data, and investigate the effect of modernising/normalising spelling and employing information extracted from a diachronic parallel corpus consisting of Dutch Bible texts from 1637 and 1977. We found that the baseline performance of a tagger trained on modern Dutch data is low (∼60%), but can be easily improved by applying a small set of rules to normalise spelling. Employing a more sophisticated method that makes use of alignments in the 17th- and 20th-century Bible versions results in an even higher within-domain score, but does not easily generalise to other 17th-century text. The best results (∼94% accuracy) on the 17th-century Bible text were achieved with a tagger trained on a corpus created by projecting the tags from the contemporary to the 17thcentury version of the Bible via automatically generated word-alignments. Adding more words to the lexicon of the tagger is an important step in generalising this result to other domains. We argue that combining the various methods discussed in the paper allows for the development of a general diachronic tagger for historical Dutch texts.