A procedure is proposed to design algorithms for intonation in synthetic speech. Its main feature is the reiterative application of two types of perceptual evaluation, followed by subsequent analyses of production data, which are guided by the results of the evaluations.