In the present bond-valence concept the bond-valence parameter ro
is treated as constant for a given pair of atoms, and it is assumed that the bond valence sij
is a function of the corresponding bond length Dij
, and that the atomic valence is an integer equal to the formal oxidation number for
Vi
derived from stoichiometry. However, from a statistical analysis of 76 [SbIIIS
n
] and 14 [SbIIISe
n
] polyhedra in experimentally determined structures, it is shown that for SbIII—X bonds (X = S, Se), ro
is correlated with {\bar \alpha}
i
, the average of the X—Sb—X angles between the three shortest Sb—X bonds. This is interpreted as a consequence of a progressive retraction of the 5s lone-electron pair from the SbIII nucleus, which can be considered as continuous change of the actual atomic valence act
Vi
of Sb from +3 towards +5. A procedure is derived to calculate an effective atomic valence eff
Vi
of SbIII from the geometry, {\bar \alpha}
i
and Dij
, of the [SbIII
Xn
] polyhedra, which approximates act
Vi
and is a better description of the actual valence state of SbIII than the formal valence for
Vi
. Calculated eff
V
SbIII
are found to vary between +2.88 and +3.80 v.u. for [SbIIIS
n
] and between +2.98 and +3.88 v.u. for [SbIIISe
n
] polyhedra. It is suggested that a corresponding modification of the present bond-valence concept is also required for other cations with lone-electron pairs.
