A sustainable route to store the energy provided by the Sun, is to directly convert sunlight into molecular hydrogen using a semiconductor performing water photolysis. Hematite (α-Fe2O3) is promising for this application due to its ample abundance, chemical stability and significant light absorption (with a band gap of 2.0 - 2.2 eV). Despite these advantageous properties, several drawbacks restrain the utilization of iron oxide for photoelectrochemical water splitting. The first limitation, namely the conduction band edge lower than the water reduction potential, can be straightforwardly overcome by adding a second solar system in tandem, which can absorb a complementary part of the solar spectrum and bring the electron at an energetic level higher than the hydrogen evolution potential. The second drawback arises from the disaccord between the short charge carrier diffusion length and the large light penetration depth. It is therefore necessary to control the hematite morphology on a length scale similar to the hole transport length. To further enhance the photoelectrochemical performance, a new concept for water splitting is introduced in this thesis. The host-guest approach consists in decoupling the different tasks of the photoanode (on one side light harvesting and water oxidation center, on the other side electron conduction to the substrate) by depositing a thin layer of hematite onto a mesoporous host (WO3 in this study). This concept has been demonstrated to increase the photocurrent by ca. 20% due to enhanced quantum efficiencies at long wavelengths. This demonstration has been nonetheless limited by the iron oxide thin films overall efficiency. Thin films photoactivity is then investigated by two means: first by controlling their nucleation on a modified substrate and secondly by incorporating plasmonic nanoparticles aimed to localize absorption in the thin film. The formation of a SiOx buffer layer on the substrate prior to deposition of hematite by Fe(acac)3 spray is shown to modify the film formation mode and its physical properties. These films exhibit photoactivity from an optical thickness of 12.5 nm (as compared to 25 nm without underlayer). The study of hematite photoanodes with gold nanoparticles, embedded or deposited on its surface, establish that charge transfer from metal nanoparticles is occurring only at overlapping wavelengths between the plasmonic resonance and the semiconductor absorption. Nevertheless, photoelectrochemical performances are reduced because of high recombination rate at the metal/semiconductor interface. Finally the third limitation, i.e. the large overpotential required to observe the onset of water splitting photocurrent, is tackled in the last part. The onset potential of photocurrent is decreased by a very thin coating of Al2O3 (0.1 - 2 nm), deposited by ALD, on the nanostructured photoanode. The subsequent application of the Co2+ catalyst further reduces the overpotential and results in a record photocurrent at 0.9 VRHE of over 0.4 mA cm-2. This investigation clearly distinguishes two causes for this energy loss: surface traps and slow oxidation kinetics. The charge accumulation and the Fermi level pinning, observed at low bias potential and assigned to these surface states were further rationalized in an investigation on photocurrent and photovoltage transients.