Gait is a new biometric aimed at recognising a subject by the way they walk. Gait differs from traditional biometrics in that it is a function of both space and time. We present two new approaches for automatic gait recognition. The first of these two approaches, area masks, aims to recognise a subject by their dynamics of area change within specific regions of the image. This approach focuses on the temporal nature of gait, which has been neglected by previous statistical approaches. The second approach, moment based descriptors, describes a shape in terms of geometric invariants. A family of shape descriptors is formed by using a masking circle proportional to the area of the image. We show that a family of descriptors is a better discriminator than using just simple moments alone. We use a state of the art database, which is currently the largest available, and present extensive experimental results examining gait as a biometric, gender discrimination, gait symmetry and performance analysis. A simple nearest neighbour classifier is used to discriminate between subjects and this provides a measure of baseline performance. Our new approaches provide promising results on the largest available database. Future work will concentrate on extending the approaches to deal with gait filmed under real-world conditions.