Biological control is a great hope for reducing the overutilization of pesticides in agricultural soils. It often involves microorganisms or molecules produced by microorganisms that will be able to interact with either a plant or pathogens of this plant to reduce the growth of the pathogen and limit its negative impact on the host plant. When new biocontrol products are developed, strains were mostly selected based on their ability to inhibit a pathogen of interest under in vitro conditions via antagonistic effects. Strains with no in vitro effect are often discarded and not tested in planta. But is the in vitro selection of bacterial agents according to their antagonism activities towards a plant pathogen the best way to get effective biocontrol products? To answer this question, we used wheat and the fungal pathogen Fusarium graminearum as a study pathosystem model. A library of 205 soil bacteria was screened in 2 types of in vitro growth inhibition tests against F. graminearum, and in an in planta experiment. We find strains which do not have inhibition phenotypes in vitro but good efficacy in planta. Interestingly, some strains belong to species (Microbacterium, Arthrobacter, Variovorax) that are not known in the literature for their ability to protect plants against fungal pathogens. Thus, developing a biocontrol product against F. graminearum must be preferentially based on the direct screening of strains for their protective activity on wheat plants against fungal diseases, rather than on their in vitro antagonistic effects on fungal growth.