In this paper, we analyse the WTO rules that specify how the subsidy provided to farmers by public stocks should be estimated. We identify three biases in these rules: - Bias B1, resulting from the use of a fixed past unit value of imports or exports as the external reference price, instead of the current price cost of imports or exports. - Bias B2, resulting from the use of the public stock procurement price instead of the price prevailing on the domestic market to estimate the price support received by farmers selling their production on the domestic market. - Bias B3, resulting from the use of total national production instead of the marketed share of national production, thereby ignoring farmers' self-consumption. The effect of these three biases on the estimated subsidy varies from country to country, but on average, WTO rules result in the subsidy being overestimated by a factor of between 2 and more than 300, depending on public stock intervention modalities and country characteristics. This means that in the most favourable scenarios, the estimated subsidy is (on average) twice the real subsidy. These biases have a huge effect on country compliance: many countries have an estimated subsidy that exceeds their maximum allowed level (even with very limited public stock interventions), simply because the subsidy provided by public stocks is overestimated by WTO rules. This result challenges the widespread idea that almost all countries comply with WTO rules on public stocks. We also test the effect of individually correcting biases B1, B2 and B3. It appears that doing so would not eliminate the bias on country compliance. One implication of this is that expressing the fixed external reference price (FERP) in US dollars, correcting it with the country inflation rate or replacing it by the average unit value of imports or exports over the last five years (as proposed by some experts and WTO Members) would not be enough to remove the bias on country compliance. It