A major problem with (partially) free constituent order is to manifest preferences among structurally distinct parses of ambiguous sentences. In order to obtain scoring criteria a preferred constituent order can considerably support a best-first strategy. This work presents an experimentally evaluated model of preferred German constituent order in the middle field and its application for the implementation of a robust and eficient parsing strategy for spontaneous speech. This constituent order is used to guide the parallel LR(1)-parser to derive the preferred interpretation of ambiguous sentences.