This paper examines the nature of the sublexical structure of signs in American Sign Language, and its relevance to various descriptive and theoretical problems. Data are presented which show that the currently accepted view of a sign as a simultaneous bundle of primes—handshape, movement, location, and orientation—is insufficient in several ways. Timing data are presented which show that the majority of signs are segmentable into movements and holds. A sequential underlying representation of signs is proposed which predicts the behavior of ‘non-contacting’ contact signs, gives a legitimate status to multiple movement signs, accounts for the three manners of movement (hold, continuous, and restrained), and allows a sequential representation of non-manual signals. It also describes in structural terms the changes which signs undergo in forming compounds, and it allows the incorporation of subject and object agreement markers into verb structure in a straightforward way. This framework allows an interesting parallel with spoken language phonology: both divide the sequential units which make up the words of the language into two major types of units. In spoken language, the units are consonants and vowels; in sign language, they are holds and movements.