The crystal structure of hydroxide perovskite Ga(OH)3, the mineral sohngeite, has been determined for a natural sample by single-crystal XRD in space group P42/nmc to R 1 = 0.031, wR 2 = 0.071, GoF = 1.208, and for comparison also in space group P42/n to R 1 = 0.031, wR 2 = 0.073, GoF = 1.076. Unit cell parameters are a = 7.4546(2) A, c = 7.3915(2) A, V = 410.75(2) A3. The two structures are very similar and both have tilt system a + a + c −. The approximate positions of all H atoms in each structure have been refined. In the P42/nmc structure all five H sites are half-occupied, whereas in the P42/n structure four sites are half-occupied and one is fully occupied. The presence of five non-equivalent OH groups in sohngeite is confirmed by single-crystal Raman spectroscopy, but does not allow a choice between these two space groups to be made. There is only a single very weak violator of the c-glide of P42/nmc and the two refined structures are essentially the same, but are significantly different from that of the original description in which orthorhombic space group Pmn21 was reported with corresponding tilt system a 0 a 0 c +. It is argued here that such a structure is very implausible for a hydroxide perovskite. On heating sohngeite to 423 K, transformation to a cubic structure with $$Im\bar{3}$$ symmetry (a + a + a +) of the aristotype occurs. This cubic phase was recovered on cooling to 293 K without back-transformation to the tetragonal polymorph. As there is no continuous group/subgroup pathway from P42/nmc (or P42/n) to $$Im\bar{3}$$ , the transformation must be first-order, which is consistent with the large hysteresis observed. The change from the tetragonal to cubic structures involves a change in tilt system a + a + c − → a + a + a +, with a significant reconfiguration of hydrogen-bonding topology. The very different tilt systems and hydrogen-bonding configurations of the two polymorphs are responsible for hysteresis and metastable preservation of the cubic phase at 293 K. As the Ga(OH)6 octahedra of the low- and high-T polymorphs are very similar it is inferred that the transformation is driven by proton behaviour, presumably involving proton re-ordering.