The problem of shape representation is a core problem in computer vision. It can be argued that shape representation is the most central representational problem for computer vision, since unlike texture or color, shape alone can be used for perceptual tasks such as image matching, object detection and object categorization. This dissertation introduces a new shape representation called the nested descriptor. A nested descriptor represents shape both globally and locally by pooling salient scaled and oriented complex gradients in a large nested support set. We show that this nesting property introduces a nested correlation structure that enables a new local distance function called the nesting distance, which provides a provably robust similarity function for image matching. Furthermore, the nesting property suggests an elegant flower like normalization strategy called a log-spiral difference. We show that this normalization enables a compact binary representation and is equivalent to a form a bottom up saliency. This suggests that the nested descriptor representational power is due to representing salient edges, which makes a fundamental connection between the saliency and local feature descriptor literature. In this dissertation, we introduce three examples of shape representation using nested descriptors: nested shape descriptors for imagery, nested motion descriptors for video and nested pooling for activities. We show evaluation results for these representations that demonstrate state-of-the-art performance for image matching, wide baseline stereo and activity recognition tasks.