In the novel domain of Data Warehouse Systems, software engineers are required to define a solution that integrates with a number of heterogeneous sources to extract, transform and aggregate data, as well as to offer flexibility to run ad-hoc queries that retrieve analytic information. Moreover, these activities should be performed based on a concise dimensional schema. This intricate process with its particular multidimensionality claims for a requirements engineering approach to aid the precise definition of data warehouse applications. In this paper, we adapt the traditional requirements engineering process and propose DWARF, a Data WArehouse Requirements deFinition method. A case study demonstrates how the method has been successfully applied in the company wise development of a large-scale data warehouse system that stores hundreds of gigabytes of strategic data for the Brazilian Federal Revenue Service.