r a 200 .021 : +61 3 n@civenv Summary This is the second of three papers describing hydrologic analyses of monthly and annual streamflow data for a global set of 729 unregulated rivers with at least 25 years of continuous data. Capacity estimates of hypothetical reservoirs are computed for each river using the Sequent Peak Algorithm (SPA), Behaviour analysis and the Gould–Dincer Gamma procedure. Based on SPA and Behaviour procedures, empirical relationships relating reservoir capacity and yield were developed which accounted for 87–96% of the variance in capacity estimates across the global data set of monthly streamflows. The theoretical Gould–Dincer Gamma procedure was also shown to be a suitable technique to estimate reservoir capacity–yield relationships. It is noted that the three procedures are based on different definitions of supply reliability. Continental variations of the estimated capacities under equivalent conditions are examined. Reservoir performance measures – reliability, resilience and dimensionless vulnerability – are computed and their continental variations described. As a result of these analyses a number of differences are noted about the performance of reservoirs across continental regions. For example, the median continental reservoir capacity as a ratio of the mean annual flow varied by a factor of 9 across the continental regions. Furthermore, based on the reliability metric as an example of reservoir performance, high reliabilities occur in the South Pacific and Europe, slightly less reliable systems in North and South America, lower still in northern Africa, followed by Australia and the lowest value in southern Africa. This distribution follows inversely with the coefficient of variation of annual streamflow between continents. a 2007 Elsevier B.V. All rights reserved.