We examined the foraging strategies of dorcas gazelles (Gazella dorcas) at different spatial scales. When in sand dunes in the Negev desert of Israel, the gazelles fed on a single plant species, the madonna lily (Pancratium sickenbergeri). In summer, when all live plant material is subterranean, the gazelles dug holes in the sand to remove the stem and bulb. After the winter rains, only the tips of the leaves that had emerged above ground were eaten. We examined search patterns and patch choice on a large scale using both null hypotheses generated by a random walk model and “rules of thumb” that may approximate on optimal foraging strategy. At a smaller scale, we examined the selection of parts of individual lilies, again making predictions about "rules of thumb" that the gazelles should use. Gazelle search paths deviated considerably from a random walk. Gazelle made shorter move lengths in areas of high plant density. Bouts of concentrated feeding in small areas were interspersed with long moves to new forging areas, suggesting that these animals were repeatedly sampling their environment. Congruent with optimal foraging predictions, the gazelles selected plants with more and larger leaves than randomly available, and concentrated their foraging activity in areas of highest lily densities. In summer, gazelles altered the depth of digs in response to differences in sand compaction. There was a surprising negative correlation between the size of the plant and the amount eaten. This selection of small plants appears to be due to the increased probability of reaching the bulb, which contains most of the plant's volume. Summer foraging by gazelles had a negative impact on the lilies. There is strong selection on the lilies to grow to a sufficient depth that damage of the bulb by herbivory is minimized. Conversely, the gazelles must select plants with bulbs close to the surface to maximize energy intake and minimize the energy cost of digging.