We briefly review the current status of the study of long-duration gamma-ray burst (GRB) host galaxies. GRB host galaxies are mainly interesting to study for two reasons: 1) they may help us understand where and when massive stars were formed throughout cosmic history, and 2) the properties of host galaxies and the localisation within the hosts where GRBs are formed may give essential clues to the precise nature of the progenitors. The main current problem is to understand to what degree GRBs are biased tracers of star formation. If GRBs are only formed by low-metallicity stars, then their host galaxies will not give a representative view of where stars are formed in the Universe (at least not a low redshifts). On the other hand, if there is no dependency on metallicity then the nature of the host galaxies leads to the perhaps surprising conclusion that most stars are formed in dwarf galaxies. In order to resolve this issue and to fully exploit the potential of GRBs as probes of star-forming galaxies throughout the observable universe it is mandatory that a complete sample of bursts with redshifts and host galaxy detections is built.