Many species have been reported to form roosting (resting, sleeping) aggregations at ‘traditional’ sites, but the alternative hypothesis that specific sites are used repeatedly because of habitat limitation is rarely tested. We studied the roosting behavior of a species of harvestman (Opiliones, Prionostemma sp.) at a lowland rainforest site in Nicaragua. Both sexes roosted by day in spiny palm trees, dispersed at dusk to forage, and rejoined aggregations just before dawn. The distribution of harvestmen among spiny palms was significantly clumped, and harvestman density did not correlate with spiny palm density. Aggregations formed repeatedly in a small subset of the available spiny palms and the same sites were used in two different years (2001, 2003). Nevertheless, the membership of aggregations was fluid; individual harvestmen were found at multiple roosts and moved up to 0.2 km per night. Translocated animals often returned to the roost where they had been released or nearby roosts but were never found at previously unused sites. The high consistency of site use but low site fidelity of individuals suggests that roost sites differed conspicuously (to the harvestmen) from sites that were not used. We found no univariate or multivariate differences between used and unused sites, however, in the characteristics of the trees or microclimate. These results conflict with the habitat limitation hypothesis but are consistent with the traditional site use hypothesis. The tradition may be mediated by a site-labeling chemical, a mechanism that does not require individual site fidelity. We discuss these results in relation to the proposed functions of roosting aggregations.