- We investigated the effects of herbivory by the dorcas gazelle, Gazella dorcas, on the production of a second inflorescence in a desert lily, Pancratium sickenbergeri (Amaryllidaceae), after the first inflorescence was eaten. In three populations exposed to different levels of herbivory, we conducted an inflorescence-clipping experiment in the flowering seasons of 1997, 1998 and 1999. The experiment was performed on inflorescences at both the emerging stage and the anthesis stage. Plants in each control group were undamaged. Clipped plants had a greater probability of producing a second inflorescence stalk than unclipped plants. In all the populations, the probability of producing a second inflorescence was greater when the first inflorescence was cut in the emerging stage than when it was cut in anthesis, indicating that there was a cost of compensation. In the population with the highest level of herbivory, the production of a second inflorescence stalk was related to resource availability in all 3 years. Plants produced as many flowers or fruits on the second stalk as on the first, indicating that potential reproductive output is double that usually achieved. Although we found no evidence for a trade-off between investment in first and second inflorescence stalks, the lower probability of producing a second inflorescence if heavy investment had been made in the first inflorescence indicates that compensation is costly. These results contradict the mutualism hypothesis that herbivory is beneficial to plants and provide support for the trade-off hypothesis that benefits of herbivory are proximal and are attained at an evolutionary cost.