Fisher's Geometric Model of Adaptation (FGM) has been the conceptual foundation for studies investigating the genetic basis of adaptation since the onset of the neo Darwinian synthesis. FGM describes adaptation as the movement of a genotype toward a fitness optimum due to beneficial mutations. To date, one prediction of FGM, the probability of improvement is related to the distance from the optimum, has only been tested in microorganisms under laboratory conditions. There is reason to believe that results might differ under natural conditions where more mutations likely affect fitness, and where environmental variance may obscure the expected pattern. We chemically induced mutations into a set of 19 Arabidopsis thaliana accessions from across the native range of A. thaliana and planted them alongside the premutated founder lines in two habitats in the mid-Atlantic region of the USA under field conditions. We show that FGM is able to predict the outcome of a set of random induced mutations on fitness in a set of Arabidopsis thaliana accessions grown in the wild: mutations are more likely to be beneficial in relatively less fit genotypes. This finding suggests that FGM is an accurate approximation of the process of adaptation under more realistic ecological conditions.