Large numerical computations, such as three-dimensional weather o~r nuclear reaction models, are an important class of codes because of the high-quality scientific results that they produce. However, they are also resource intensive in that they can require the use of multiple processors and/or multiple levels of memory; they are often both computationally intensive and generate a very large amount of data. Partitioning of grid data for efficient transfer among multiple processors or multiple levels of memory may be a key element in the design of efficient codes for large numerical computations. A survey of data organization in such codes is presented, and partitioning schemes that were used are classified. Three classes of partitioning are described, and the relationship between numerical method and data organization is explained. Design strategies and implementation languages for partitioning provide the scientist with tools for code development.