The principal source of streamflow during periods of low flow in the Susquehanna River basin of New York is the discharge of groundwater from sand-and-gravel deposits. Spatial variation in low flow is mostly a function of differences in three watershed properties: the amount of water that is introduced to the watershed and available for runoff, the extent of surficial sand and gravel relative to till-mantled bedrock, and the extent of wetlands. These three properties were consistently significant in regression equations that were developed to estimate several indices of low flow expressed in cubic feet per second or in cubic feet per second per square mile. The equations explain 90 to 99 percent of the spatial variation in low flow. A few equations indicate that underflow that bypasses streamflow-measurement sites through permeable sand and gravel can significantly decrease low flows. Analytical and numerical groundwater-flow models indicate that spatial extent, hydraulic conductivity and thickness, storage capacity, and topography of stratified sandand- gravel deposits affect low-flow yields from those deposits. Model-simulated discharge of groundwater to streams at low flow reaches a maximum where hydraulic-conductivity values are about 15 feet per day (in valleys 0.5 mile wide) to 60 feet per day (in valleys 1 mile wide). These hydraulic-conductivity values are much larger than those that are considered typical of till and bedrock, but smaller than values reported for productive sand-and-gravel aquifers in some valley reaches in New York. Differences in the properties of till and bedrock and in land-surface slope or relief within the Susquehanna River basin of New York apparently have little effect on low flow. Three regression equations were selected for practical application in estimating 7-day mean low flows in cubic feet per second with 10-year and 2-year recurrence intervals, and 90-percent flow duration, at ungaged sites draining more than 30 square miles; standard errors were 0.88, 1.40, and 1.95 cubic feet per second, respectively. Equations that express low flows in cubic feet per second per square mile were selected for estimating these three indices at ungaged sites draining less than 30 square miles; standard errors were 0.012, 0.018, and 0.022 cubic feet per second per square mile, respectively.