During the last decade, many studies have addressed the signature of community assembly processes by assessing the deviation of a functional diversity pattern from that expected in null communities, using a randomization algorithm. The basic principle has been to compare the functional structure of a ‘local community’ with a set of randomly generated communities from a ‘regional species pool’. Specifically, it allows assessing the extent to which species constituting a community are functionally more or less similar than expected under the assumption of a random assembly, thus revealing the influence of assembly processes such as competition or environmental abiotic filtering. However, the processes shaping community assembly act at various scales, and their combined influences on the functional community structure may be difficult to disentangle. In this context, we need a framework that explicitly takes into account the multi-scale nature of the ecological processes in order to perform appropriate analyses and draw unambiguous interpretations. By combining a comprehensive conceptual scheme and a review of the literature, we have showed that performing a trait-based null model analysis cannot be reduced to the sole choices of a randomization algorithm and a functional metric. Other choices, mainly related to data acquisition also appear crucial, from the sampling strategy, i.e. the choice of an appropriate scale, to the type of data, i.e. mainly presence-absence versus abundance, through the proper definition of a reference species pool. Because the methodological choices are non-independent, we found that particular combinations of modalities of these methodological choices seem more suitable to detect environmental filtering, competitive dominance or limiting similarity patterns. In light of these findings, we propose some practical guidelines to test the influence of the community assembly mechanisms at multiple spatial scales.