Irrigation systems are critical to agricultural systems in semi-arid parts of the developing world. Although there is ample evidence that canal systems fail to reach their design capacity, there have been surprisingly few studies of the allocation efficiency of water within canal systems. Partly this is due to poor data concerning water withdrawals per farm. In the first part of this study, we collected refined measures of water withdrawals and find evidence supporting the hypothesis that farmers near the head of a canal get more water than farmers near the tail. Accounting for the conveyance efficiency of the canal system ameliorates the efficiency loss somewhat. The analysis builds a strong evidence-based case that water is not allocated efficiently now within the canal. The results suggest that improvements in canal water management or an internal water market would yield efficiency gains for the canal. In the second part of this study, we analyse farmer adaptations to heterogeneous canal water availability. Farmers adapt to reduced flows by reducing their overall planted area. Next, they modify their crop mix by switching from a water intense crop (cotton) to a crop that is less sensitive to water (millet). Finally, we consider input choice and find, not surprisingly, that most inputs are complementary to surface water irrigation and reductions in surface water deliveries result in reductions in use of other inputs. We explore two cases more thoroughly. First, we find that own-labour tends to increase as canal water decreases and we test to find that this tends to be a function of scale. Finally, we consider an input of special interest, groundwater, which we expected to act as a substitute to surface water. Instead, we see evidence of complementarity to surface water. This suggests that groundwater quality plays a distinct role in its usage and we do find evidence of groundwater quality modulating the amount of groundwater usage in tandem with surface water use.