Construction of a regional flood frequency curve is based, as a rule, on fitting this curve to representative quantiles. In a regional sample of floods the probability of extreme values corresponding to return periods, that exceed the record lengths, is much larger than that of individual series, used to determine the representative quantiles. The probabilities of exceedance of regional extremes can be calculated straightforward in case of independent data, applying the theory of order statistics. For regionally dependent data one can define an equivalent number of independent regional series and then utilize the theory for independent data. This approach is exemplified with flood data from Norway.