In sign language studies, it is generally assumed that a sign can be divided into several phases in time (preparation, stroke, and retraction) and that the stroke contains all of the necessary information. However, this has not been tested empirically. In order to learn where the information truly resides, we present an experiment that investigates the distribution of information in a sign. Signers were shown isolated fragments of Dutch Sign Language signs (citation form) and were then asked to identify the sign. The results show that the stroke alone performs as well as the entire sign. However, the preparation, together with the transition from preparation to stroke, produces equally good recognition, thereby suggesting that most of a sign's information is available early. Surprisingly, in many cases preparation alone and retraction alone also produce quite good recognition (66 percent and 60 percent, respectively). The recognition pattern across signs gives an indication of signers' recognition strategy. (Contains 8 figures and 2 tables.)