Normalised resistance interpretation was applied to the estimation of epidemiological cut-off values from data sets that contained > 150 observations from fully susceptible wild-type (WT) strains for six bacterial groups/antibiotic combinations. A series of 20 subsets containing 10, 20, 30 and 40 WT observations were randomly generated from these six larger data sets. The epidemiological cut-off values estimated using these 480 small data sets were compared with those estimated from their larger, parental data sets. Eighty-seven per cent of the 360 epidemiological cut-off values estimated from subsets that contained a parts per thousand yen20 WT observations were in agreement with those estimated from their respective larger sets. The remaining 13 % differed only by one doubling dilution. These calculations suggest that small strain sets can be used to set epidemiological cut-off values for MIC data without unacceptable loss of precision. It is concluded that the design of experiments intended to generate the data necessary to set epidemiological cut-off values should aim to include observations on at least 30 WT observations.