Requirements for a machine description facility for compiler testing are discussed. The compiler testing procedure consists of proving that programs are correctly translated by the compiler at hand. This is achieved by use of a common intermediate representation for both the source and object programs. The intermediate representation for the object program is built by use of a process termed symbolic interpretation. This process interprets a set of procedures which describe the effects of machine language instructions corresponding to the target machine on a suitable computation model in a manner consistent with an execution level definition of the high level language. Some of the important factors which enter into such a definition are discussed. These include architectural constraints posed by the target machine, and a description facility for memory and data types. Once such a definition is formulated, the actual instruction set of the target machine can be described. The highlights and limitations of such a definition facility are discussed in the context of a specific LISP implementation on a PDP-10 computer.