In this paper we introduce a new method for the synthesis of Spanish prosody suitable for the automatic generation of prosody in a Text-to-Speech system. As some methods we have already proposed our approach is data-based, it models joint F0 contour and segmental durations and its linguistic analysis is rule-based. Unlike previous works it uses a mixture of a priori breath group classification (linguistically based) and data-based phonological mapping. This new approach together with the previous ones form a quite open framework for analysis and synthesis of Spanish prosody. This approach leaves room for growing from a general prosodic model towards application specific prosody. The new method is successfully tested in a particular style of telephone number reading, quite dificult to pick up by previous methods.