In a study of farmer knowledge and experimentation of trees and tree cultivation, two categories of farmer consultants were identified, tree experts (n = 44) and control farmers (n = 70), via a two stage process of 'knowledge' ranking and community interviews conducted in three communes in southern Rwanda in 1992. A detailed inventory of tree species was conducted revealing two key characteristics for on-farm trees: multiple utility and high locational flexibility. In order to logically organize the species inventory, two indices, a utility index (UI) and a locational flexibility index (LFI), were developed. The two indices were calculated for each species and then plotted as the axes on a bivariate grid. While few species had both high utility and flexibility indices, the majority that did were indigenous species long used in agroforestry systems (e.g., Euphorbia tirucalli, Vernonia amygdalina, Erythrinu abyssinica, etc.). A few naturalized exotics introduced in the 1930s (e.g., Grevillea robusta, Eucalyptus camaldulensis, E. saligna, E. maideni, Cupressus lusitanica and Persea gratissima) were also appreciated for their multiple utility and locational flexibility. Farm sizes varied, averaging 1.27 ha among tree experts and 0.18 ha in the control group. The mean number of species cultivated was significantly higher on tree expert farms (P < 0.01). However, tree density was much lower on tree expert farms (731 trees-ha J) than on control farms (1,689 trees.ha-~). Competitive species were found primarily on larger farms, whereas smaller farms were associated with species of greater UI and LFI. Farmers demonstrated deliberate and rational choices about which trees to plant in accordance with each species utility and flexibility in relation to land availability.