The petroleum industry has been a strong consumer of traditional supercomputers, routinely running technical applications which process terabytes of data and require Gflops-years of computation.Now that comparable parallel supercomputers can be assembled using commodity components for about one-tenth the price of a traditional supercomputer, it is inevitable that the petroleum industry will adopt commodity systems.In this paper, we will first give an overview of industrial seismic imaging for petroleum exploration, including the physics and computational approach of the predominant algorithm.We then will describe our decision-making process for and experience in constructing a commodity supercomputer.This machine has been performing production seismic imaging since January 1, 1999.