To face increasing uncertainties, future farming systems must be sustainable not only under average conditions but also in extreme climatic and economic situations. Various concepts such as stability, robustness, vulnerability or resilience have been proposed to analyze the ability of agricultural systems to adapt to changing production conditions. The operational effectiveness of these concepts remains nevertheless limited. In this paper, we developed an original analytical framework allowing characterizing and quantifying crop yield robustness, as well as identifying agricultural practices linked to cropping systems differentiated according to their robustness pattern. This framework was applied to 2300 bread wheat plots belonging to 145 cropping systems in various regions of France over the period 2011-2014. The analysis was performed at the scale of the cropping system. In a first step, we defined a regression statistical model allowing us to link wheat yield variability to an index of abiotic perturbations constructed using the STICS agronomic model; the cropping systems were taken into account through the use of dummy variables. In a second step, the different cropping systems were positioned within four quadrants using the regional average wheat yield in conditions of average abiotic perturbations and the regional average estimated robustness to abiotic perturbations as cut-offs for the quadrants. In a third step, the cropping systems of the different spaces defined by the four-quadrant approach were compared on the basis on three types of agronomic practices, i.e., management intensification, rotation and heterogeneity practices. Empirical results show that abiotic perturbations had an impact on wheat yield variability. This impact differed from one system to another which means that there is a "cropping system effect" of abiotic perturbations on wheat yield robustness. Several agronomic practices allowed differentiating high versus low wheat yield cropping systems. High yield cropping systems relied more intensively on chemical inputs (fertilizers and pesticides) and used more diversified rotations, with more frequently legumes as preceding crops and a lower frequency of cereals. Fewer agronomic practices allowed differentiating robust versus sensitive wheat cropping systems. In addition to the sowing date (later for robust systems) and the sowing density (greater), these practices were essentially linked to spatial adjustments of the sowing date, total pesticide use, variety earliness at heading stage and variety disease resistance.