The interactions between the design of a computer's instruction set and the design of compilers that generate code for that computer have serious implications for overall computational cost and efficiency. This article, which investigates those interactions, should ideally be based on comprehensive data; unfortunately, there is a paucity of such information. And while there is data on the use of instruction sets, the relation of this data to compiler design is lacking. This is, therefore, a frankly personal statement, but one which is based on extensive experience. My colleagues and I are in the midst ofa research effort aimed at automating the construction of productionquality compilers. (To limit the scope of what is already an ambitious project, we have considered only algebraic languages and conventional computers.) In brief, unlike many compiler- compiler efforts of the past, ours involves automatically generating all of the phases of a compiler-including the optimization and code generation phases found in optimizing compilers. The only input to this generation process is a formal definition of the source language and target computer. The formulation of compilation algorithms that, with suitable parameters, are effective across a broad class of computer architectures has been fundamental to this research. In turn, finding these algorithms has led us to critically examine many architectures and the problems they pose. Much of the opinion that follows is based on our experiences in trying to do this, with notes on the difficulties we encountered.