Sweden experienced a number of large floods in the 1980-ies. This raised the question of whether floods were becoming more frequent. A systematic study on floods was carried out, to provide a perspective to past and future floods. Frequency analysis was made using 16 methods. A split-sample test was used for evaluation of the predictive power of the methods. Numerical criteria were used for measuring the goodness of fit. The return periods of observed floods were estirnated by use of plotting positions.

No convincing evidence of trends was found. The 1980-ies had larger floods than usual, where as the 1970-ies had few high floods, especially, in the auturnn. This may have led to the, irnpression of a trend. No evidence of autocorrelation or periodicity was found. For most of Sweden, the spring was found to be the season with highest extremes. The flood-moderating effect of regulation was illustrated, although floods were also found to occur in regulated systems. The results from the frequency analysis depended on the choice of criteria. In general, however, distributions with two parameters performed best. Neither a two component model treating spring and autumn separately, three parameter distributions, nor regional analyses gave any irnprovements. The spatial correlation within the data was considerable.