In a supertagging task, sequence labeling models are commonly used. But their limited ability to model long-distance information presents a bottleneck to make further improvements. In this paper, we modeled this long-distance information in dependency formalism and integrated it into the process of HPSG supertagging. The experiments showed that the dependency information is very informative for supertag disambiguation. We also evaluated the improved supertagger in the HPSG parser.