In this paper, the efficient encoding and manipulation of large sets of two-dimensional data that represent multiple non-overlapping features is investigated. Both linear structures are examined, i.e., leafcode and treecode representations of thematic layers. A new technique is presented and evaluated by applying it to already proposed access methods that are based on the previous representations. More specifically, we experiment with window queries that involve multiple features having as a performance measure the number of disk accesses. Experimentally it is shown that treecode representations outperform the previous structures with respect to time and space complexity.