The Warwick Pyramid Machine is a large parallel architecture for image processing and computer vision applications. It consists of several distinct processing layers, each well suited to a particular class of operations: a very fine grain array of numeric processing elements, one per pixel; an intermediate array of scalar processors, capable of providing independent local control of regions within the image; and a loosely coupled array of powerful, general purpose processors for symbolic manipulation. The authors describe their current implementation of this architecture, in the form of both software simulators and hardware prototypes combining the DAP and transputers, and show how the machine is suited to performing iconic to symbolic data transformation. A number of representative algorithms are described, including low-level signal processing, feature extraction, segmentation and classification, as well as applications in the field of graphics. >