This paper proposes a highly-parallel two-dimensional cellular automata architecture called CAM2 for real-time and palm-top pixel-level image processing and presents results of its performance evaluation. CAM2 can attain pixel-order parallelism on a single board because it is composed of a CAM (Content Addressable Memory), which makes it possible to embed an enormous number PEs (Processing Element), corresponding to CA (Cellular Automaton) cells, onto one VLSI chip. Multiple-zigzag mapping of a CA cell to a PE and dedicated CAM functions enable high-performance CA processing. The performance evaluation results show that 256k CA cells, which correspond to a 512 × 512 picture, can be processed by a CAM2 on a single board using deep sub-micron process technology. The processing speed is 10 billion CA cell updates per second under a four-neighbor condition. CAM2 will make a significant contribution to the development of compact and high-performance systems, especially real-time and palm-top pixel-level image processing systems.