The study considers the seasonal distribution of non-tidal peak flows on a large number of rivers draining varied catchments across Scotland and Northumberland. Peaks over threshold (POT) flood series from 156 gauging stations are used, and are subject to two quality control measures. Firstly, threshold values are standardised to give 45 peaks over a ten year period and secondly, records are adjusted to compensate for non-stationarity in the sampled data. The database assembled consists of 3458 station-years of record. A comprehensive description of the seasonal patterns found is presented, based on these quality-controlled data and utilising a number of methods of characterisation. Directional statistics are employed to indicate the central tendency of time-of-year values for each station, a six-season analysis gives more detailed information, and the seasonality of large peaks is compared with that of full POT series. Finally, a classification analysis is used to summarise these patterns. These patterns are related to five catchment characteristics: the seasonality of rainstorms; soil moisture deficit lengths; catchment size; lake storage and snowmelt, although the effect of the last of these is unclear as suitable data were not available for analysis. A discriminant analysis is employed to relate the five physical factors to flood seasonality. The study concludes with a discussion on the implications of its findings. A method of assessing seasonal flood risk using POT series is presented, offering an accurate means of relating flood magnitude to recurrence interval for any period of less than one year. The implications of seasonal heterogeneity, both within and between flood records, are also discussed. The suitability of the exponential model for use with POT records is questioned and it is suggested that explicit recognition of the seasonality of flooding may be necessary in order to make accurate design flood estimates.