Water is one of the major limiting factors in vegetation recovery and reconstruction in the extremely arid desert region of northwest China. We established five moisture gradients to simulate a desert habitat, and measured the photosynthetic light responses of two-year-old Populus euphratica seedlings under drought stress using the Li-6400 portable photosynthesis system. The resulting data were fitted and analyzed using rectangular hyperbola, non-rectangular hyperbola, exponential and modified rectangular hyperbola models. This study aimed to optimize these models, investigate the adaptability of light-response models to different levels of drought stress, and understand P. euphratica adaptation to arid desert conditions. The results showed the net photosynthetic rate (P_n) decreased with increased drought stress, and the total P_n declined under the same photosynthetically active radiation (PAR). Relatively high photosynthesis rates and reduced photo-inhibition were observed in P. euphratica under conditions of low-moderate drought stress (relative soil water content (RSWC) > 45%). When models were used to express this, the exponential, rectangular hyperbola, and non-rectangular hyperbola models fit well with the P_n and PAR response process, but the maximum net photosynthetic rate (P_nmax) were higer than the measured values and the light saturation point (LSP) were lower than the measured values (P < 0.01). When the RSWC was lower than 45% (where seedlings would experience greater drought stress), and when an obvious photo-inhibition and significant decrease in P_n were observed under strong light, LSP and P_nmax were also significantly decreased. Under these circumstances, only the modified rectangular hyperbola model could fit the light response processes and light response parameters. Overall, the goodness of fit of the four light response models to the observed data was as follows:modified rectangular hyperbola model > exponential model > non-rectangular hyperbola model > rectangular hyperbola model. The adaptability of the four models to different levels of drought stress was variable. The modified rectangular hyperbola model perfectly simulated the light response process and fitted photosynthetic parameters under all water conditions. It was especially suited to severe drought stress and extremely arid desert conditions, whereas the other three models were only suitable under high moisture conditions. The response threshold of the light response parameters differed with the changing levels of drought stress. The apparent quantum yield (AQY), P_n, P_nmax, and LSP of P. euphratica decreased gradually with increasing levels of drought stress, but its dark respiration rate (R_d) and LCP increased significantly under extremely severe drought stress. P. euphratica showed higher AQY, P_nmax, and LSP under RSWCs greater than 45%, while its P_nmax and LSP decreased significantly when RSWC was less than 45%. The photosynthetic efficiency of P. euphratica decreased during drought stress, as did its light tolerance range. Under severe drought stress, photo-saturation and photo-inhibition increased significantly, while photosynthetic capacity and normal seedling growth were substantially inhibited. P. euphratica was sensitive to small changes in drought severity and decreased its light use efficiency and respiration to adapt to desert conditions. We conclude that to enhance species conservation and restore vegetation in extremely arid desert regions, the soil water content should be maintained at about 50% of the field capacity, which would optimize desert plant growth and water management in the Tarim Desert area.