A new parallel architecture that is amenable to optical implementation is presented for massively data-parallel computing. The architecture is an SIMD model that exploits spatial parallelism and processes 2-D binary images as fundamental computational entities. Processing is based on a new technique called symbolic substitution logic. A hierarchical mapping technique is presented for designing data-parallel algorithms and mapping them onto the optical architecture. The mapping of several numerical algorithms onto the architecture is presented. Implementation issues as well as performance analysis of the optical system are also considered.