A field experiment was conducted in France to evaluate the impact of health information on fish consumption. A warning given to the treatment group revealed the risks of methylmercury contamination in fish and also gave consumption recommendations. Using difference-in-differences estimation, we show that this warning led to a significant but relatively weak decrease in fish consumption. However, consumption of the most contaminated fish did not decrease despite advice to avoid consumption of these types of fish. Accompanying questionnaires show that consumers imperfectly memorize the fish species quoted in the warning. The results point to the relatively poor efficacy of a complex health message, despite its use by health agencies around the world.