Data flow is a mode of parallel computation in which parallelism in a program can be exploited at the fine grained as well as macro level. A data flow computer executes a data dependency graph rather than the program counter controlled sequence of instructions executed by conventional machines. Nonprocedural languages appear to be especially appropriate high level languages for data flow computers. Nonprocedural languages have only two statement forms: data description and assertion. The assertions enumerate the relationships among the data. A data dependency graph is also a suitable representation for a nonprocedural language program (or specification). This research is concerned with translating the dependency graph form of a specification to a program graph for a data flow machine. Specifications in the MODEL language are translated into an intermediate form, the data flow template. The template is a language-independent representation of the specification. The template is then translated into a data flow language (Manchester Dataflow) for the Manchester University machine. The translation consists of creating an array graph to represent the specification; generating the data flow program template from the array graph; and translating the template into MaD.