The precarious condition of the world's fisheries is making ever-greater demands of the scientific assessment of fish stocks. Traditional assessments that rely on commercial catch statistics can have major shortcomings1 (as shown, for example, by the collapse of Canada's northern cod stock2), increasing the need for more fishery-independent data. Acoustic surveys can provide such information3, but ocean-going research vessels have high operating costs, and there is also widespread concern that fish avoid these vessels because of the noise they make, thereby biasing abundance estimates4. Here we present new data gathered by an autonomous underwater vehicle (AUV) showing that vessel avoidance is not a significant source of bias. Our investigation also heralds the arrival of AUVs as effective survey platforms.