We investigated the effects of role distribution on individuals' choice of reference frames in a two-person task. Pairs of participants had to move a virtual block in a constraint immersive virtual environment: only one of them could manipulate the ob-ject, his coworker guided him in the VE. Results show that the guiding operators used more addressee-centered frames of ref-erence than the manipulators. They also suggest that the guides tried to facilitate the manipulators' actions by endorsing the cog-nitive load through the production of spatial utterances.