Equations developed by multiple-regression analysis of data from 49 drainage basins in Massachusetts, New Hampshire, Rhode Island, Vermont, and southwestern Maine indicate that low flow of streams in this region is largely a function of the amount of water available to the basin and the extent of surficial sand and gravel relative to the extent of till and fine-grained stratified drift. Low flow per square mile from areas of surficial sand and gravel is consistently much greater than that from areas of till and bedrock, but flood plains and alluvial fans seem to contribute less low flow per square mile than do other types of surficial sand and gravel. The areal extent of lakes and swamps also correlates negatively with low flow in multiple-regression equations, presumably because intense evapotranspiration from these localities consumes water that would otherwise become streamflow. The annual minimum 7-day mean low flows that occur during summer and fall at 2-year and 10-year recurrence intervals (7Q2 and 7QIO) were selected as indices of low flow and were adjusted to a common base period, 1942-71. Central New England was divided into a region of high relief that com- prises much of New Hampshire, Vermont, and western Massachusetts, and a region of low relief that generally lies to the east and south but also includes the Lake Champlain lowland of Vermont. In the high-relief region, mean basin elevation proved to be the most significant index of the amount of water available. In the low-relief region, mean annual runoff per square mile was more significant than elevation, particularly when multiplied by the areal extent of sand and gravel and that of till. Dividing the areal extent of sand and gravel by stream length improved the fit of regression equations for the low-relief region. Regression equations were developed that explained at least 95 percent of the variation in 7QIO within both the high-relief and the low-relief data sets. Equations proposed for practical application were reasonably consistent with the statistical assumptions of least-squares analysis and yielded 7Q2 and 7QIO values with standard errors of 1.9 and 1.4 cubic feet per second, respectively, for the high-relief region and 2.2 and 1.6 cubic feet per second for the low-relief region. When error was expressed as a percentage of each observed value, median errors were about 25 percent for 7Q2 in both regions, and about 25 and 55 percent for 7QIO in the high-and low-relief regions, respectively. The equations do not apply to basin segments that are substantially affected by urbanization, stream regulation, or ground-water withdrawals, and may not be appropriate where basin characteristics fall outside their range in the data set or where the geologic and topographic maps needed for measurement of basin characteristics are unavailable, or are of small scale or mutually inconsistent.