Floods are common hazards which recur at various frequencies and intensities. For rational planning of water resources development, the risks of floods should be determined in probabilistic terms. Many parts of the world have a sparse hydrometric network, however, and the records are often not very long. In this paper, two approaches that extract salient information from short-term records are presented and a method that estimates the probability distribution of flood characteristics for ungauged basins is also given. Studies were made using northern Ontario as an example. The experience gained in analysing the floods of this region can be applied profitably to those countries where hydrological data are limited.