It has often been observed in the literature that when a language allows for scrambling definite noun phrases may freely scramble whereas indefinites are subject to certain restrictions. Many have assumed that scrambling is related to the structure of the surrounding discourse such that familiar or topical (anaphoric) noun phrases must scramble whereas new or non-topical (non-anaphoric) are not allowed to scramble. We will show, however, that in similar discourse configurations certain indefinites are not allowed to scramble, whereas definites freely scramble. We account for the difference in scrambling behaviour between definites and indefinites within a version of what might be called Optimality Theoretic Semantics (OTS). Our version of OTS gives a new twist to the strategy of Partee and Rooth (1983) to assign NPs a preferred type–often the simplest type possible–by using the preferences to phrase soft constraints. In this way one obtains a means to judge correct type-theoretic derivations as more or less optimal, and to vary the grammaticality judgment of the output accordingly. This is in stark contrast with the older strategy, where the output of correct derivations on the basis of unpreferred types would still be judged grammatical. More in particular our explanation of scrambling behaviour will be along the following lines. Different types of NPs come with different basic word orders, a variation which is consistent with the frequency of use of these constructions. Deviation of basic word order yields a shift in meaning. Definites, unlike indefinites, allow for an interpretation in the scrambled and in the unscrambled position without change of meaning. Accordingly, scrambling of definites is truly optional. Scrambled indefinites, on the other hand, are either infelicitous or may induce a change in interpretation; they cannot be optional. This is especially clear for inputs consisting of an (indefinite with a light verb. Then, definites are free to scramble, whereas indefinites cannot scramble. We show how the interpretive tendencies that arise with scrambling can be modelled using a notion of optimal derivation, phrased in terms of the preferred types of (in)definites and verbs. Two appendices provide the technical details of the semantic insight on which our approach is based.