Fourier Transform InfraRed (FTIR) spectroscopy is an increasingly used technique in biology, especially for whole cell metabolomic fingerprint. The reproducibility of this technique is influenced by a large number of factors such as the physiological state of cells, sample manipulation and growth conditions. Evidence exists suggesting that the cell shape and dimension can be further elements to consider in whole cell FTIR analysis. In this study we aimed to address the effect of cell geometry on the FTIR spectra and to define the extent of variability occurring between machine and biological replicas with a standardized protocol. The yeast species Saccharomyces cerevisiae (large oval-shaped cells) and Debaryomyces hansenii (small round shaped cells) were employed for their different morphology. Thirty machine replicas of each were analyzed separately and after averaging in groups of three, showing a three to four-fold reduction of the variability. Similarly, a two-fold reduction of variability was observed when thirty biological replicas of the two yeast species were analyzed. The optimal number of replicas to average was then estimated with a bootstrap-like procedure in which biological and machine replicas were randomly resampled 2000 times and averaged in groups spanning from 2 to 12 replicas. This simulation has shown that little if any advantage can be obtained by increasing the number of replicas over five and that the variability exhibited by the small regular cells of D. hansenii was always roughly half of that displayed by the large S. cerevisiae cells, confirming the results obtained with standard non-bootstrapped averages.