Abstract Background To date, the decision to set up therapeutic extra-corporeal life support (ECLS) in hypothermia-related cardiac arrest is based on the potassium value only. However, no information is available about how the analysis should be performed. Our goal was to compare intra-individual variation in serum potassium values depending on the sampling site and analytical technique in hypothermia-related cardiac arrests. Methods Adult patients with suspected hypothermia-related refractory cardiac arrest, admitted to three hospitals with ECLS facilities were included. Blood samples were obtained from the femoral vein, a peripheral vein and the femoral artery. Serum potassium was analysed using blood gas (BGA) and clinical laboratory analysis (CL). Results Of the 15 consecutive patients included, 12 met the principal criteria, and 5 (33%) survived. The difference in average potassium values between sites or analytical method used was ≤1 mmol/L. The agreement between potassium values according to the three different sampling sites was poor. The ranges of the differences in potassium using BGA measurement were − 1.6 to + 1.7 mmol/L; − 1.18 to + 2.7 mmol/L and − 0.87 to + 2 mmol/L when comparing respectively central venous and peripheral venous, central venous and arterial, and peripheral venous and arterial potassium. Conclusions We found important and clinically relevant variability in potassium values between sampling sites. Clinical decisions should not rely on one biological indicator. However, according to our results, the site of lowest potassium, and therefore the preferred site for a single potassium sampling is central venous blood. The use of multivariable prediction tools may help to mitigate the risks inherent in the limits of potassium measurement. Trial registration ClinicalTrials.gov Identifier: NCT03096561.