Central to any legacy migration project is the translation of the data model. Decisions made here will have strong implications to the rest of the translation. Some legacy languages lack a structured data model, relying instead on explicit programmer control of the overlay of variables. In this paper we present our experience inferring a structured data model in such a language as part of a migration of eight million lines of code to Java. We discuss the common idioms of coding that were observed and give an overview of our solution to this problem.