Based on runoff records for the period 1938–48, peak‐rate probability curves are computed for 51 watersheds located in the Allegheny‐Cumberland Plateau and ranging in size from 100 to 350,000 ac. A simple correlation is computed to determine the relationship between the peak rate for a ten‐year recurrence interval and area. The average standard error was 34.8 pct. Two rainfall factors representative of the period of runoff record are then introduced together with a topographic factor, and a multiple correlation is computed. The average standard error was reduced to 18.2 pct. Long‐term rainfall factors are then computed and used in the multiple correlation equation to determine the peaks that would have resulted from the peak‐rate probability curves had long‐term runoff records been available. Relationships between ten‐year peaks and those for 25‐ and 50‐year frequencies are determined from the peak‐rate probability curves and the Plateau is divided into two zones based o