Troponin measurement is central to the management and risk stratification of acute coronary syndromes. Decisions are made by categorizing troponin as positive or negative. We sought to evaluate categorical agreement between four troponin assays. Sixty blood samples were analysed by three troponin I assays (Centaur, Architect and point-of-care i-STAT) and one troponin T (TnT) assay (Roche Elecys). The upper reference limit was taken as the lowest value with a coefficient of variation of 10% or less. Continuous agreement between assays was good (Pearson's correlation coefficient 0.871–0.995). Categorical agreement assessed by Cohen's kappa varied from poor (between Architect and Centaur κ = 0.37, and between TnT and Centaur κ = 0.48) to good (between Architect and i-STAT κ = 0.68, and between TnT and i-STAT κ = 0.68). Percentage of positive results varied almost twofold, from 37% for the Centaur to 72% for the Architect. Comparison of four troponin assays showed up to twofold variations in the proportion of positive results. This implies that either a large proportion of troponin-positive diagnoses are missed by some assays or the assays with higher positivity are generating large numbers of false positives. Clinicians should evaluate troponin results in the clinical context and not base decisions solely on the ‘normal range’ of their local troponin assay.