An assessment of the performance of an EC 1441/2007 microbiological criterion was conducted based on the new Poisson-gamma modelling framework characterising within-batch and between-batch variability in Enterobacteriaceae counts on pre-chill sheep carcasses. Since the model does not assume within-batch constant variance but instead represents an association between within-batch means and dispersion measures, the operating characteristic (OC) curves could be constructed with confidence intervals arising from the uncertainty in the within-batch spread conditional to the within-batch mean. The model predicted that in Ireland the microbiological criterion would categorise the hygiene of sheep processing as ‘satisfactory’ (below mT = 1.5 log CFU/cm2) and ‘acceptable’ (between mT and MT = 2.5 log CFU/cm2) on average 98.6% (95% CI: 84.6–100%) and 1.4% (95% CI: 0–14.8%) of the times a batch is tested by taking the mean log of five individual samples. On average, batches produced beyond ∼260 CFU/cm2 and ∼1660 CFU/cm2 of mean Enterobacteriaceae concentration would have at least 95% confidence of being spotted by the ‘acceptable’ and the ‘unsatisfactory’ criterion (>MT), respectively, although under the existing contamination levels virtually no tested batch will prompt the revision of hygiene procedures in the Irish sheep abattoirs. This work proposes the definition of microbiological limits in arithmetic means, as by simulation this approach was found to lead to the definition of sampling plans that are both more effective (i.e., reduced uncertainty around the acceptance probabilities) and with more discriminatory power than those based on the common mean log scale. Based on the relatively low levels of Enterobacteriaceae on Irish pre-chill sheep carcasses (4.5 CFU/cm2), a new lower limit mT′ of 60 CFU/cm2 derived from the arithmetic means approach has been suggested so that batches of up to the 95th percentile of mean concentrations (∼22 CFU/cm2) be considered as originating from a ‘satisfactory’ process with at least 95% confidence. This more conservative mT′ implies that 1.5% (95% CI: 0–16.3%) of the tested batches will emit a warning signal in the long run. Increasing sample size however will not affect the average probability of warning but will have an impact on reducing the uncertainty resulting from the between-batch variability.