A study of the relationship between grain yield (GY) and grain protein concentration (GPC) in bread wheat was carried out on a 11-year series of trials conducted by the Groupement d'Etude des Varietes Et Semences (GEVES) for the registration of new cultivars on the French National List. Values for GY and GPC came from 458 individual trials, grouped together in 21 series of bi-annual and multi-site data. The correlations between the two characters, calculated environment by environment, appeared highly variable due to high "genotype x environment" interactions for GY and GPC. The use of mean values, calculated on the 21 series of GEVES trials, enabled a better assessment of the relationship between the two characters, and an algorithm was proposed to avoid bias due to potential outliers. Using the well-assessed relationship obtained, grain protein deviations (GPDs) were defined as the standardized residuals of the regression of GPC on GY. These deviations appeared to have a partly genetic basis, as the lines with high deviations were about the same in the two independent datasets constituted by the two consecutive years of GEVES experiments. Some lines used as standards in GEVES trials obtained significant GPD for different series of bi-annual and multi-site trials, confirming the genetic origin of high-GPD. Simulations made to determine the minimum experimental design, showed that at least five sites per year for two consecutive years, were necessary to have a good assessment of the GY-GPC relationship, and hence reliable estimates of GPD