Agronomic performance of wheat genotypes is usually evaluated by studying the mean value of pertinent agronomic variables. In particular, grain yield is analyzed through its different average numeric components (number of plants per m-2, number of spikes per plant, number of spikelets per spike, number of grains per spike, number of grains per m-2, individual grain weight). Nevertheless, the variance of these components has not be so far fully explored. Moreover, number of grains per m-2 is often considered as the most important component impacting grain yield. Yet it has been observed that the bigger the number of grains is, the more the number of small grains is, especially those in apical spikelets or in apical position within a spikelet. The variance associated to the grain size and individual grain composition (nitrogen concentration) is linked to both tillering and spike structure. Its impact on agronomic performance, especially in stressful environments (abiotic stress) is not much known. The aim of this study is to analyze the genetic and ecophysiological determinism of wheat individual grain characteristics variability and evaluate its impacts on agronomic performance in order to quantify the actual genetic variance potential. To this aim, several issues have to be investigated: (i) understand precisely the structure of the variance associated to individual grain characteristics and (ii) understand causal processes of this variance, especially spatial structure of the spike and spikelets. The present work is based on a field experiment where 228 genotypes were grown under optimal conditions or under a severe water stress. All yield components were recorded. Out of this panel, four genotypes were selected for in depth analysis of grain traits variance structure. For these genotypes, 10 spikes were randomly chosen in each plot and were individually dissected in order to record the exact position of each grain within the spike structure. Grains originating from 2 spikes among these 10 spikes, nitrogen concentrations per grain were also measured. Our results show that (i) individual grain weight is strongly affected both by its position within the spike as well as by mean spike characteristics and (ii) variance components are under both genetic and environmental control. These results suggest that studying variable components of grain traits can help defining typologies of wheat genotypes differing by their strategies to set grain number (and thus potential yield) at anthesis. Further studies will be needed in order to test whether these strategies have differential effects on wheat tolerance to abiotic stresses during grain filling.