In 2005, we published a paper (Legendre et al. 2005; hereafter referred to as LBP) explaining the ecological and statistical basis for the analysis of the variation in species composition among sites (this is one operational definition of beta diversity), a subject of great impor tance for the understanding of the generation and maintenance of beta diversity and the establishment of ecologically sound ecosystem conservation policies. Tuomisto and Ruokolainen (2006) (hereafter referred to as TR) pointed out several issues in our 2005 paper. As a response, this note seeks to clarify the issues behind this contention. Our major point of disagreement concerns the links between (1) the ecological predictions derived from neutral theories of beta diversity, (2) the statistical hypotheses derived from these predictions, and (3) the statistical methods used to test these hypotheses. In LBP, TR, and the present Comment, distance-based methods refer to statistical methods where the geo graphic relationships among sites are represented by a distance matrix. These methods include the Mantel test and the derived method of regression on distance matrices. The raw-data approach refers to multiple regression and canonical analysis, where the spatial relationships among sites are represented by a rectan gular table containing geographic coordinates, a poly nomial of the geographic coordinates, principal coordinates of a neighbor matrix (PCNM) (Borcard and Legendre 2002), or derived forms. The purpose of this Comment is to show that (1) some predictions of Hubbell's neutral theory, especially the presence of positive autocorrelation, can be stated and tested using the raw-data approach, (2) the distance approach as proposed by TR (partitioning on distance matrices) is statistically flawed, and (3) when a raw-data hypothesis is translated into distances, the correspond ing statistical test lacks power; therefore, whenever a hypothesis can be formulated in terms of raw data