The orthographic surface structure of Swedish words has been used for predicting parts-of-speech information using a connectionist approach. This technique can be used to aid syntactic processing within a text-to-speech system. The error back-propagation technique has been used for the connectionist learning. A corpus of the 10 000 most frequent Swedish words have been used for training and testing the system. The results indicate that around 80% of the words can be correctly classified by using the last part of each word. The system is compared to a rule based system that makes the same sort of predictions from word endings. Both systems give comparable results for the lexicon used.