A　hydrological　characteristic　-based　methodology　for　dividing　flood　seasons　into　sub-seasons　is　proposed　to　make　full　use　of　a　reservoir＇s　flood　control　storage　during　the　flood　season　via　multi-stage　flood-limited　water　levels.　The　proposed　framework　is　mainly　composed　of　three　parts:　the　selection　of　indices　depicting　flood　seasonality,　the　establishment　of　segmentation　methods　that　can　address　clustering　problems　with　high-dimensional　time　series　and　unknown　numbers　of　clusters,　and　multi-scheme　comparison　and　rationality　analysis.　The　reasonability　and　validity　of　the　proposed　framework　is　illustrated　through　an　empirical　case　study　of　China＇s　Panjiakou　Reservoir　basin.　The　results　indicated　that　the　dynamic　fuzzy　c-means　method　with　clustering　validity　function　provided　more　objective　and　quantitative　divisions　than　other　methods,　including　the　Fisher　optimal　partition.　The　flood　season　of　the　Panjiakou　Reservoir　basin　(June　1-September　30)　is　divided　into　three　sub-seasons　according　to　the　principle　of　optimal　clustering:　a　pre-flood　season　(June　1-July　10),　a　main　flood　season　(July　11-August　20),　and　a　post-flood　season(August　21-September　30).