A computerized system for three-dimensional tracking of large numbers of individual free-flying insects was used to assess the performance of Drosophila melanogaster from populations that had undergone 160 generations of selection for upwind flight ability. Compared with control lines, the selected lines showed significant increases in mean flight velocity, decreases in angular trajectory and a significant change in the interaction between velocity and angular trajectory. Maximal flight velocity was apparent as a sharply defined upper boundary of the distribution of horizontal and vertical velocity as a function of angular trajectory; this upper bound (0.85 ms-1) differed little between the selected and control lines, although individuals from the selected lines attained maximal performance levels much more frequently. Maximum induced power output was calculated directly from the product of maximum vertical velocity and body weight. This measure (28 W kg-1 muscle) was closely predicted by a scaling relationship derived from the load-lifting limits of larger insects and vertebrates, as well as tethered D. melanogaster stimulated via their optomotor reflex to produce maximal lift. These results indicate that selection for flight performance can readily alter the relative effort and/or the frequency of phenotypes capable of attaining population-wise maximal performance levels, but shows little ability to increase population-wise maximal performance.