A vowel sound can be characterized acoustically in terms of its formant pattern, and generally by the first three formants. However, all combinations of F1, F2, F3 are not produced by human speakers. Vowel space is contained in what we call the Maximal Vowel Space (MVS), which is well known and even analytically approximated in the F1-F2 plane. The aim of this work is to determine acoustic and articulatory respective contributions to characteristics related to the MVS. From a 4-tube model, we precise intrinsic acoustic phenomena which limit the F1-F2 realization domain ; then, in order to bring out the specific effects of articulatory constraints, we explored the MVS with the help of two models which integrate, in their way, certain fundamental knowledge of speech production. Differences between F1-F2 maximal spaces generated in these three cases are acoustically discussed and explained in relation to articulatory constraints and knowledge of vowel production. Finally, a tridimensional description of the MVS and a parameterization in the F1-F2 plane as a function of F3? is proposed with the help of a model that best reflects articulatory reality.