Abstract: Characteristics of giant panda herbivory sites and clonal regeneration of a bamboo Fargesia robusta following giant panda grazing were studied in Anzihe Nature Reserve, Sichuan Province of southwestern China. Three types of plots were randomly selected in a panda'summer habitat in 2002: herbivory (naturally grazed by giant pandas), clipped (simulated panda herbivory), and control. Then, the clonal population regeneration dynamics was monitored and compared in following three years (from 2009 to 2011) in these three sites. The results demonstrated that: 1) the features of panda herbivory plot were as follows: pandas avoided eating thin bamboo culms with basal diameter 7.5 mm; the mean density ofgazed bamboo rod was (10.872.85)individuals/m2; the mean height of gazed clums was (0.790.10) m; the rate of panda gazing was (22.203.15)%; and the closest mean distance between the trees (DBH 10 cm) near herbivory sites was (0.670.29) m. 2) In 20092011, without panda gazing, there was no significant difference between the annual rates of recruitment and mortality and no significant differences between the rates in varied years either. The F. robusta population was keeping a dynamic balance. Herbivory increased can enhance the rate of mortality, however, it also stimulated the rate of recruitment at the same time. Especially, the annual rates of recruitment was significantly higher than the mortality rate (P =0.000) one year after herbivory. Panda herbivory can promote regeneration in first three years from feeding activity, and three years later, F. robusta will return to normal regeneration rate and maintain the dynamic balance. There was no significant difference between new living bamboos after panda herbivory and controls. The interference of panda herbivory to bamboo forest will not affect the future usage of bamboos. Clonal population has compensatory effect on panda herbivory. 4) In the clipped plot, the rate of recruitment was significantly higher than that of mortality (P2009 =0.001) in 2009; but no significant difference between them was observed in 2010 and 2011 (P2010 = 0.928; P2011 = 0.538), which was in consistent with the clonal population dynamics in the herbivory plot. This suggests that manual simulation can effectively mimick natural panda herbivory and promote the clonal regeneration.