Methodology for sensory profiling of fermented milks has been studied. A vocabulary of sensory attributes has been developed which encompasses most of the difference between samples. Principal Component Analysis, incorporating factor rotation, allowed simplification of the variability between samples to five Principal Components capable of clear interpretation. Sensory mapping was found to be a useful tool for categorizing fermented milks. The acceptability of the fermented milks was successfully modelled, by Partial Least Squares Regression, in terms of a limited number of key attributes. The model explained 88.4% of the variance. The relations between sensory attributes and the composition of the fermented milks were considered using Multiple Linear Regression. Although a number of statistically significant relations were derived they were of poor to modest value for purposes of prediction.