To fulfill the requirement of rapid access to huge amounts of uncompressed pixmap image data, a parallel image server architecture is proposed, based on arrays of intelligent disk nodes, with each disk node composed of one processor and one disk. It is shown how images can be partitioned into extents and efficiently distributed among available intelligent disk nodes. The image server's performance is analyzed according to various parameters such as the number of cooperating disk nodes, the sizes of image file extents, the available communication throughput, and the processing power of disk node and image server processors. Important image access speed improvements are obtained by image extent caching and image part extraction in disk nodes. With T800 transputer-based technology, a system composed of eight disk nodes offers access to three full-color 512×512 pixmap image parts per second (2.4 megabytes per second). For the same configuration but with the recently announced T9000 transputer, image access throughput is eight images per second (6.8 megabytes per second)