Eighty-seven two-hour acoustic surveys (radius 0.8 nautical mile, vertical range 0–500 m) around 17 fish aggregating devices (FADs) were conducted in French Polynesia between December 1995 and February 1997. Associated tuna densities were calculated using two different techniques: echo counting when the fish had sufficient distances from each other and echo integration when the fish swam close together (in schools). No acoustic detection of tuna was observed during 27 of the 87 surveys, representing 81 % of all the nocturnal surveys and 15 % of the diurnal ones. The 60 other surveys showed three different classes of aggregations: (1) ‘deep scattered fish’, observed 45 times, (2) ‘intermediate scattered fish’, observed 16 times, and (3) ‘shallow schooling fish’, observed 16 times. Sometimes aggregations of different classes were observed beneath the same FAD. The size of the fish inside the aggregations (determined from target strength values), the distance between the individuals, and the depth of the fish all decreased from ‘deep scattered fish’ to ‘shallow schooling fish’ (100–300 m for ‘deep scattered fish’, 50–150 m for ‘intermediate scattered fish’, and above the depth of 50 m for ‘shallow schooling fish’). Fish densities also varied according to the class of aggregations: 7.3, 26, and 801 fish per km3 on average for ‘deep scattered fish’, ‘intermediate scattered fish’, and ‘shallow schooling fish’, respectively. The highest densities were observed during daytime, while night-time observations indicated a variety of situations, from the absence of individuals to large amounts of fish.