A large number of database index structures have been proposed over the last two decades, and little consensus has emerged regarding their relative e_ectiveness. In order to empirically evaluate these indexes, it is helpful to have methodologies for generating random queries for performance testing. In this paper we propose a domain-independent approach to the generation of random queries: choose randomly among all logically distinct queries. We investigate this idea in the context of range queries over 2-dimensional points. We present an algorithm that chooses randomly among logically distinct 2-d range queries. It has constant-time expected performance over uniformly distributed data, and exhibited good performance in experiments over a variety of real and synthetic data sets. We observe nonuniformities in the way randomly chosen logical 2-d range queries are distributed over a variety of spatial properties. This raises questions about the quality of the workloads generated from such queries. We contrast our approach with previous work that generates workloads of random spatial ranges, and we sketch directions for future work on the robust generation of workloads for studying index performance.