Aims: Analyse the controversial relationship existing between wood density, precipitation, temperature and aridity, and its importance in shaping forest communities facing drought. Location: New Caledonia (SW Pacific). Methods: We sampled wood density in 1580 trees belonging to 175 species across nine sites (eight 1-ha plots) located in dry, mesic and humid forests. We tested whether wood density varied across species, communities and habitats, and depended on mean annual precipitation (MAP), mean annual temperature (MAT) and an aridity index (AI). We also tested whether the variations observed in community mean wood density (WDmean) and in wood density standard deviation differed from those obtained by a randomized species distribution across communities. We constructed a phylogenetic supertree at genus level and tested for the non-random distribution of WDmean. Results: We found that WDmean tended to increase with increasing aridity (i.e. decreasing MAP and increasing MAT) and was significantly higher than expected under randomized species distribution at the dry forest site only. However, some species with low wood density grew well at these dry sites. At genus level, wood density was not random within the phylogenetic supertree. Some lineages exhibiting higher wood density than expected by randomization were also the most represented at the dry sites. Conclusions: Although we did not observe an environmental filter sensu stricto, aridity associated with a low MAP and a high MAT at the dry and mesic forest sites was likely to favour species with a high wood density. Our phylogenetic results suggested that species belonging to a few lineages exhibiting high wood density were likely to be adapted to drought, hence favoured by increasing drought. (Résumé d'auteur