In this paper we present a metrical-prosodic analysis of Dutch read aloud sentences, based on perceived prominences. Metrical-prosodic constraints are formulated which can be used as input for Text-to-Speech systems. A four level metrical grid representation is introduced, corresponding to four degrees of prominence, including no prominence. A distinction is made between input and output constraints. The former refer to the prosodic representation of lexically distinguished categories; the latter to sentence-level prosodic well-formedness. Constraints may be in conflict with each other. The conflicting constraints are relatively ranked in a constraint hierarchy. The higher ranked constraint will win at the cost of the lower ranked one.