Currently, there is general concern about the non-stationary behaviour of flood series. Consequently, several studies have been conducted to identify large-scale patterns of change in such flood series. In Spain, a general decreasing trend was found in the period 1959-2009. However, a multi-temporal trend analysis with varying starting and ending years showed that trend signs depended on the period considered. Flood oscillations could influence the results, especially when flood-rich and flood-poor periods are located at the beginning or end of the series. In Spain, a flood-rich period in 1950-1970 seemed to lead to the generalised decreasing trend, as it was located at the beginning of the flood series. Nevertheless, the multi-temporal test can only find potential flood-rich and flood-poor periods qualitatively. A methodology has been developed to identify statistically significant flood-rich and flood-poor periods. The expected variability of floods under the stationarity assumption is compared with the variability of floods in observed flood series. The methodology is applied to the longest streamflow series available in Spain. Seven gauging stations located in near-natural catchments with continuous observations in the period 1942-2014 are selected. Both annual maximum and peak-over-threshold series are considered. Flood-rich and flood-poor periods in terms of flood magnitudes and the annual count of exceedances over a given threshold are identified. A flood-rich period in the beginning of the series and a flood-poor period at its end are identified in most of the selected sites. Accordingly, a flood-rich period placed at the beginning of the series followed by a flood-poor period influence the generalised decreasing trend in flood series previously found in Spain.