Most models of habitat selection assume that individual animals choose and either reuse or abandon sites based on a constant reassessment of site quality. When survival is a function of the presence of conspecifics, however, the benefits of returning to traditional sites may override resource assessment. Many animals form roosting aggregations at what appear to be traditional sites. At our study site in Nicaragua, the harvestman Prionostemma sp. forms diurnal roosting aggregations on a small subset of the available spiny palm trees. With respect to physical characteristics and microclimate, the spiny palms used by the harvestmen resembled a random sample of those available, yet the same subset of trees was used in two different years (2001, 2003). This suggests that the location of aggregation sites is traditional, not a product of habitat limitation. Individual harvestmen were not faithful to particular roost sites, however, which raises the question of how the tradition could be maintained over time. In this paper, we present evidence, derived from a series of small-scale field experiments, that the harvestmen mark roosting sites chemically and enter marked sites preferentially when searching for places to roost. We also show that the harvestmen are sensitive to changes in site quality (the presence of spines) but will continue to use degraded traditional sites when no intact spiny palms are nearby. This system provides an example of how animal traditions could be maintained over multiple generations without learning. Site-labeling can be viewed as an external form of social memory.