This paper presents an adaptation of Fujisaki's quantitative intonation model to German and its application to F0 synthesis by rules. The parameter values of the model were determined by an automatic approximation of naturally produced F0 contours. By means of statistical methods, the sources of variation of the parameter values were examined. On the basis of the significant factors, standard values were derived that capture the effects of both linguistic and speaker-dependent features. A set of rules was formulated in order to generate intonational prototypes that relate to linguistic categories such as sentence modality, sentence accent, and word accent. By means of the TD-PSOLA algorithm, the original F0 data of utterances were replaced by the rule-generated contours. Acceptability of the artificial intonation patterns as well as the adequate realization of linguistic categories were evaluated perceptually by berth phonetically trained subjects and 'prosodically naive' listeners. In the first experiment, both groups of listeners were able to tell the original contours and their close approximations apart, although presumably mainly on the basis of some unavoidable audible effects of the F0 manipulation and not by perceived differences in the intonation contours. As it turned out in the second experiment, the rule-generated contours for declaratives were generally not much less acceptable than the close approximations of the original contours, while the ratings for interrogatives were significantly lower. The experts' ratings were considerably higher than those of the naive subjects. The third experiment aimed at obtaining detailed judgments regarding the realization of word accents and sentence modality that will help to improve several specific rules.