Aims Shellfish production areas are classified for suitability for human consumption using counts of Escherichia coli in shellfish samples. Two alternative laboratory methods are approved in the European Union and UK for measuring E. coli in shellfish samples; the most probable number (MPN) and pour plate methods. These methods have inherently different statistical uncertainty and may give different counts for the same sample. Using two approaches: simulated data and spiking experiments, we investigate the theoretical properties of the two methods to determine their reliability for shellfish waters classification. Methods and results Assuming a Poisson distribution of E. coli in shellfish samples, we simulate concentrations in 10 000 samples using the MPN and pour plate methods. We show that for higher concentrations of E. coli the pour plate method becomes increasingly more reliable than the MPN method. The MPN method has higher probabilities than pour plate of generating results exceeding shellfish classification thresholds, while conversely having higher probabilities of failing to detect counts that exceed regulatory thresholds. The theoretical analysis also demonstrates that the MPN method can produce genuine extreme outliers, even when E. coli are randomly distributed within the sampled material. A laboratory spiking experiment showed results consistent with the theoretical analysis, suggesting the Poisson assumption used in the theoretical analysis is reasonable. Conclusion The large differences in statistical properties between the pour plate and MPN methods should be taken into consideration in classifying shellfish beds, with the pour plate method being more reliable over the crucial range of E. coli concentrations used to determine class boundaries.