Ecologists use approaches based on plant functional traits to tackle several fundamental and applied questions. Although a perfect characterization of functional trait structure requires the measurement of all the individuals in communities, this is prohibitively resource-consuming. Consequently, the general practice is to average the trait values of a reduced number of individuals per species. However, there are different alternatives regarding the number, identity and spatial location of the individuals chosen to calculate species-averaged trait values. In this study, we compared different strategies for sampling functional traits, using community-weighted mean trait values (CWM) and the Rao index of functional diversity (FD). We intensively sampled the functional trait structure along a topographical gradient in a Mediterranean grassland, obtaining accurate estimations of the ‘real’ values of these indices (CWMI and FDI) for three traits (height, leaf area and specific leaf area). We simulated three different sampling strategies differing in the spatial location of the individuals used to estimate species-mean trait: i) average of the whole gradient (GLO), ii) average of the sampling unit in which the abundances of species maximize (MAX) and iii) average of a reduced number of individuals per species and sampling unit (LOC). For each strategy, we simulated different sampling intensities (number of individuals sampled). For each trait, we examined the ability of each strategy and sampling intensity to accurately estimate CWMI and FDI, as well as their ability to detect changes in functional trait structure along the topographical gradient. LOC outperformed the other strategies in terms of accuracy and bias, and was much more efficient to describe changes along the gradient, regardless of the traits and indicators considered. Furthermore, LOC was the only strategy that improved consistently as sampling intensity increased, especially at low levels of intensity. Our results indicate that the impact of considering intraspecific variability in trait values can be greater than commonly assumed. Strategies that neglect this source of variability can result in inaccurate or biased estimations of the functional trait structure of plant communities. Most importantly, we show that intraspecific variability can be taken into consideration without any increases in the total number of individuals measured.