Strong support for a ''mixture'' model of liquid water can be found fr
om an analysis of the accurate experimental density data (H2O) over th
e range t similar to -30 degrees C in the supercooled regime to t simi
lar to +70 degrees C. Published density data can be fit to this mixtur
e model with six- to seven-decimal-point precision. Remarkably, the ou
tput parameters from these fits indicate the presence of capacious int
ermolecular bonding with a density extremely close to that of ordinary
ice-Ih, intermixed with compactly bonded regions having a density nea
r that of the common dense forms of ice, in particular ice-II. Densiti
es at higher temperatures could also be fit to good precision with suc
h a model, though the model must clearly become less valid as the temp
erature rises and more varied bonding forms contribute. The fitting pr
ocedure also shows that both the capacious and dense components have p
ositive thermal expansion coefficients that are similar in magnitude t
o those of their respective ice forms. As T approaches the vicinity of
225 K in the deep supercooled regime, the structure of the liquid app
roaches disordered ice-I-type bonding, with no contribution from the d
ensely bonded component. Combined with the differential X-ray scatteri
ng data of Bosio, Chen, and Teixeira on liquid water, and structural d
ata on the ice polymorphs from Kamb's work, it can be concluded that t
he bonding differences between the dense and capacious structures are
not at the nearest-neighbor level but occur instead in the outlying no
n-hydrogen-bonded next-nearest-neighbor O...O structure. Because of th
e long-range structural implications of this conclusion, uncertainties
arise in molecular dynamics modeling of the liquid and on the usefuln
ess of attempts to learn about the liquid from the study of small gas-
phase clusters.