Statistical analysis of long-term changes in marine communities generally involves recourse to multivariate methods. Some of these, such as Correspondence Analysis (CA), are very sensitive to the presence of rare species in the data, whereas other methods, such as Principal Components analysis (PCA) or Multidimensional Scaling (MDS) are only sensitive to the dominant species. Thus, it is necessary to identify rare species. We used an original method of species determination, based on a multinomial scheme and permitting the generalization of the classical presence/absence coding from a probabilistic viewpoint. This selection method was applied to the identification of the different categories of species (rare, intermediate or dominant) that constitute the muddy fine sand community of the Bay of Morlaix. Multidimensional Scaling was performed on the Bhattacharrya interdistance table between observations, based on dominant species. The results of this analysis were compared to those of several correspondence analyses carried out on both tables of presence/absence codings (the usual and the generalized ones).