The spring flood is a multivariate event that can be characterized by the floodcharacteristics peak discharge, duration and volume. Traditionally these flood characteristics have been modelled separately, but by modelling thespring flood as a multivariate event, information about the joint probabilistic behaviour is obtained. In this thesis the bivariate logistic extreme value distribution with GEV margins are used to model asymptotically dependent flood characteristics, and the bivariate log normal distribution with log normal margins are used for modelling asymptotically independent flood characteristics.A new method is made for the assessment of the durationof the spring flood. The bivariate models and the method for assessment of duration are tested in the Glomma basin in the rivers Glomma and Gudbrandsdalslågen at stations Elverum and Losna, respectively. At Elverum the method for assessment of duration gave reasonable results, whereas at Losna the method gave poor results. For the bivarate models there were good agreement between the models and observations, and the models were found appropriate for modelling the corresponding flood characteristics. The bivarate joint distributions of the flood characteristicswere used for constructing bivariate return periods.