The choice of the sediment size fraction in the analysis of fossil planktonic foraminifera is of great importance in determining the composition of assemblages. In past studies several size fractions have been utilised. Imbrie and Kipp (1971) stated that “smaller fractions give rise to too many uncertainties in the identification of small specimens and require too long to process. Large mesh size yield undesiderable loss of small species, and small specimens of larger species”. The adoption of coarser sieve meshes has the effect to reduce the percentages of small sized (usually living in cold waters) species (Bé and Hutson, 1977). In a comparison between >63 μm and >150 μm size-fraction of planktonic foraminifera assemblages from NW Atlantic Ocean, Smart (2002), stated that because particular smaller species are either under-represented or even absent from the larger (>150 μm) size-fraction, the smaller (>63 μm) size-fraction must be included in studies of planktonic foraminifera. Di Donato et al. (2008) highlighted in the >150 μm size fraction of a core from the Tyrrhenian Sea, a great loss of the small-sized species Turborotalita quinqueloba in glacial samples, where this species is very abundant. This caused an increase in warm water species such as Globigerinoides ruber, apparently reaching typical Holocene values in some full glacial levels. From the above mentioned statements, it is clear that treatment changes can strongly influence the results of palaeoclimatic reconstructions based on planktonic foraminiferal assemblages. As stated by Aitchison (1986, 1992) scale invariance and subcompositional coherence are fundamental properties of the compositional data analysis (CODA). The main goal of this paper is to verify if the variable relationships pointing out from CODA of foraminiferal assemblages are or not influenced by the size fraction and, more in general, to test the robustness of CODA respect to treatment changes related to different preparation techniques for the analysis of planktonic foraminifera