A high performance architecture for EBCOT in JPEG 2000 is proposed. The architecture consists of a column-and-pass dual parallel context modeling unit capable of the simultaneous processing of four samples, and a pipelined binary arithmetic coder based on a two-context window running at double the clock frequency to match the processing rate of the context modeling unit. Simulation results show that this design reduces processing time by about 60% compared with existing pass-parallel architectures and is able to process a 512/spl times/512 grayscale image in approximately 4 ms at 100 MHz system clock rate.