An algorithm is described for synthesizing intonation in the absence of syntactic information as part of an experimental system for text-to-speech conversion for Dutch. It contains several improvements of an existing algorithm. The improvements are based on analyses of speech data collected from a professional reader. They mainly concern the intonational realization of intonation phrases. The performance of the algorithm has been evaluated in a perceptual test, in which listeners were asked to judge the naturalness of the intonation. For isolated utterances, the rule-based intonation is judged more natural than the intonation generated with the older algorithm, and as natural as the human intonation. For coherent text, the rule-based intonation is judged more natural than the intonation generated with the older algorithm, but less natural than the human intonation. These findings suggest that the decomposition of the utterance into intonation phrases by means of pitch makes an essential contribution to the naturalness of synthetic speech. In addition, they support earlier findings, reported in the literature, which point to the perceptual relevance of text intonation.