Matalas et al. (1975) (MSW) found that the simulated values of flood p
eaks in a region using various common forms of flood frequency distrib
utions did not reproduce the empirical skew statistics for 14 differen
t regions covering the conterminous United States. Specifically, the f
ield data always demonstrated a higher value of standard deviation of
skew for a given sample value of mean skew than did the simulations. M
SW termed this difference ''separation'' and further showed that it co
uld not be explained either by autocorrelation of flood peaks or as a
small sample property. In this paper, we discuss an explanation of thi
s property using the recently developing scaling theories of regional
floods. It is shown that in a homogeneous region, recently defined by
us, separation would result from the multiscaling structure of flood p
eaks. Separation would not occur if floods obey simple scaling, nor wo
uld separation necessarily occur with heterogeneity or mixing among di
fferent homogeneous regions. Mixing must be of a particular kind in or
der to cause separation. The use of normalized flood frequencies havin
g mean of zero and variance of 1 in the simulations carried out by MSW
is shown to be consistent with the assumption of index flood or simpl
e scaling but not multiscaling. In the 14 ''megaregions'' analyzed by
MSW, mixing among subregions within each megaregion may add to the mag
nitude of separation. The separation in 14 regions is physically inter
preted based on different physical mechanisms that have been recently
hypothesized by us to be responsible for the presence of simple scalin
g or multiscaling in floods.