Photo-induced water splitting using semiconductor photocatalysts has attracted considerable attention for producing H2 as a clean energy carrier, while the effective utilization of visible light is imperative to achieve the desired efficiency for practical applications. However, it turned out that it is essentially difficult to develop metal oxides that possess a bandgap below 3.0 eV for visible light absorption while placing a conduction band minimum (CBM) more negative than the water reduction potential, because the valence band maximum (VBM) of conventional metal oxides are fairly positive due to the dominant contribution of O-2p orbitals. Thus, mixed-anion compounds such as oxynitrides have been intensively studied as candidates since one can expect that higher energy p orbitals of non-oxide anions (e.g., N-2p) elevate their VBM values. Unfortunately, most of them are subject to facile self-oxidation by photogenerated holes, while highly dispersed cocatalyst particles certainly improve the stability of some oxynitrides. We have recently demonstrated that Sillén–Aurivillius type perovskite oxyhalides such as Bi4NbO8Cl can stably and efficiently oxidize water to O2 under visible light without any surface modifications, and also exhibits a stable Z-scheme water splitting when coupled with a H2-evolving photocatalyst. It was revealed that the VBMs of these materials consist mainly of O-2p orbitals, instead of Cl-3p (or Br-4p), but their positions are much more negative than those of conventional oxides. Since O– anions are known to be relatively stable, photogenerated holes populated at the O-2p orbitals will not lead to self-decomposition but to oxidize water. These results could provide new strategies for developing durable photocatalytic materials for water splitting under visible light, by manipulating the interaction between post-transition metal s orbitals and O-2p orbitals.