Starting from a previously published stoichiometric model for the commensurate Type III phase in the (1–x)Bi2O3∙xNb2O5 system, Bi94Nb32O221 (x = 0.254), we have developed a crystal-chemical model of this phase across its solid-solution range 0.20 ≤ x ≤ 0.26. After using annular dark-field scanning transmission electron microscopy to identify the metal sites that support non-stoichiometry, we show that the maximum possible range of that non-stoichiometry is 0.198 ≤ x ≤ 0.262, perfectly consistent with the experimental result. Inter-site cation defects on these sites provide some local coordinative flexibility with respect to the surrounding oxygen sublattice, but not enough to create continuous fluorite-like channels like those found in the high-temperature incommensurate Type II phase. This explains the reduced oxide-ionic conductivity of Type III compared to Type II at all temperatures and compositions, regardless of which phase is thermody-namically stable under those conditions. The solid-solution model shows that oxygen disorder and vacancies are both re-duced as x increases, which also explains why Type III becomes relatively more stable, and why oxide ionic conductivity decreases, as x increases.