Fishers have exploited the associative behavior displayed by several pelagic fish species with floating objects for decades, through the use of man-made fish aggregating devices (FADs), which facilitate the capture of such species. However, our understanding of this associative behavior and its adaptive value is poor and the scientific community is ill-equipped to provide fishery managers with science-based recommendations on the impacts of FADs on ecosystems. In an array of 13 anchored FADs around Oahu, Hawaii, USA, 72 yellowfin tuna Thunnus albacares were equipped with internal acoustic tags, which facilitated the continuous monitored of their presence and absence around each FAD using automated acoustic receivers. Data were analyzed using survival curves with the objective of determining the behavioral dynamics of fish joining and leaving the FADs. Residence times at FADs were characterized by 4 behavioral modes: briefly passing near a FAD (average 13.1 min), short association (average 2.9 d), and 2 long association behaviors (13.8 and 23.2 d, respectively). Statistical analyses suggest that different behavioral modes were likely dependent upon local conditions around the FAD at a given time (environmental factors or social interactions). We observed 2 behavioral modes for absence times from FADs: short (2.8 d) and long (infinite). More importantly, individuals exhibited behavioral variability, switching between short and long residence times at FADs. This suggests that large pelagic fish can display a range of behavioral responses while in an array of FADs, challenging the common hypothesis of a single behavioral pattern, which could ultimately lead to an ecological trap. Survival curves were best fitted with exponential models, suggesting that underlying behavioral processes were time independent.