A new approach to constructing a prosodic grammar of Mandarin read speech which describes the mapping from syntactic patterns to prosodic patterns is proposed. It first prepares a large read-speech corpus with syntactic-tree parsing for texts and break-index labeling representing four-layer non-recursive prosodic hierarchical structures for utterances. Then, all realizations of syntactic pattern-break pattern pairs are extracted to learn prosodic grammatical rules. For a syntactic pattern, rules are inferred via calculating the break-type distributions of pre-boundary, post-boundary, and intra-pattern word junctures from these realizations. In the study, we only investigate the prosodic grammatical rules for four syntactic patterns of determinative-measure (DM) compound, DM+N, DM+DE+N and DM+Modifier+DE+N to verify the feasibility of the proposed approach. With considering the ten syntactic functions of Subject, Object, Topic, Head, Modifier, Attributive, Noun Predicate, Quantitative Complement, and embedded in DE phrase and in Prepositional Phrase, the entropies of pre- and post-boundaries of these four syntactic patterns are reduced significantly. Moreover, detailed rules are inferred via exploring linguistic/semantic interpretations for the occurrence of main prosodic pattern and outliers using the information of phonetic constituents and contexts of syntactic-pattern realizations. Some important factors such as length and semantic relation are found to seriously affect the syntax-prosody mapping.