Altered water regime of reservoirs has changed the interactions among native plant species, thus leading to many serious environmental problems. To restore the riparian vegetation is of great significance. Among many plant interaction studies, the stress-gradient hypothesis has been supported by many researchers, while some research did not support it. The stress-gradient hypothesis indicates that the interactions among plants largely depend on external stresses, and the importance of facilitation is predicted to increase with the severity of environmental stress. In this research, we had two purposes: (1) to explore the responses of the biomass and morphology of Cynodon dactylon to different water treatments and different plant densities; (2) to test whether these responses support the stress-gradient hypothesis. Four water treatments including normal water supply (CK), light flooding followed by light drought (FD), light flooding (LF) and flooding (FL) were applied to C. dactylon seedlings. For each water treatment, four densities including 1, 2, 4, and 12 plants per pot were used as the planting treatment. Biomass and morphological traits were determined at final harvest. The intensity of intraspecific competition of C. dactylon was determined by the relative neighbor effect (RNE). The results showed that: (1) the dry biomass of plant leaf, stem, root, aboveground biomass, root:shoot ratio and total biomass of C. dactylon were significantly affected by water treatments (P < 0.001). The biomass of C. dactylon was reduced with increasing water stress regime. The different densities and the interaction between soil moisture and density also significantly affected the leaf, stem and root dry mass, aboveground biomass and total biomass of C. dactylon (P < 0.001). (2) Area per leaf, branching number, total stem length and internode length of C. dactylon were significantly affected by water treatments (P < 0.001), and the different densities and the interaction between soil moisture and density significantly affected branching number, total stem length and internode length, but had little effect on the area per leaf. (3) Different water treatments significantly affected the value of RNE of biomass, and these results supported the stress-gradient hypothesis. The negative values of RNE of biomass in CK and LF indicated that the relationship among plants was competition, while the positive values in FL suggested a facilitative relationship. (4) The RNE values of total stem length were negative, but increased with the intensity of water stress at middle and high density level, which showed that the relationship among C. dactylon changed from competitive to facilitative with increasing water stress. This may be due to adjacent plants helping to reduce the negative effects of water stress on their neighbours. To sum up, the impact of water and density stress on morphological characteristics of C. dactylon was less than that on the biomass. From this research, we can conclude: (1) The survival rate of C. dactylon was 100%, and C. dactylon showed morphological responses that could alleviate its water stress. In riparian zones, C. dactylon is a suitable species for ecological restoration, while identifying the best planting density needs further research. (2) The growth and morphology of C. dactylon showed some negative influence of increasing planting density. (3) The stress-gradient hypothesis was supported by our research. However, refinement of this conceptual model is needed to improve it.