Previous algorithms to compute lexical chains suffer either from a lack of accuracy in word sense disambiguation (WSD) or from computational inefficiency. In this paper, we present a new lineartime algorithm for lexical chaining that adopts the assumption of one sense per discourse. Our results show an improvement over previous algorithms when evaluated on a WSD task.