Objective: Thyroid disease can be subtle in its presentation, and TSH reference intervals may be artefactually increased by including persons with subclinical thyroid disease. We have therefore used a thyroid disease‐free population to determine TSH and fT4 reference intervals. Design: Apparently healthy subjects were assessed by health questionnaire, drug history, clinical assessment and measurement of thyroid antibodies. Patients: Healthy subjects in a community setting. Measurements: TSH, free T4, antithyroglobulin and anti‐TPO were measured on the Abbott Architect analyser. Subjects with clinical abnormalities, consumption of thyroid‐active medications or with thyroid antibodies above the manufacturer‐quoted reference intervals were excluded. TSH and fT4 data were log‐transformed, and the central 95% was used to calculate reference intervals. We assessed whether these data were normally distributed. We compared samples spanning the reference intervals for both TSH and fT4 between different assays looking at biases. Results: From a population of 1,606 subjects, 140 males (18%) and 284 females (34%) were excluded. The central population 95% for TSH was 0·43–3·28 mU/l and for fT4 10·8–16·8 pmol/l. There were no age‐ or sex‐related differences. For both analytes, the distribution was not significantly different to a Gaussian distribution (P > 0·05). For 5 commonly used assays for TSH, the maximum difference in the upper limit of the TSH reference interval was 0·48 mU/l and for fT4 the maximum difference for the upper reference limit was 4·1 pmol/l. Conclusions: A substantial proportion of apparently healthy persons have subclinical thyroid disease. These subjects must be excluded for any thyroid hormone reference interval studies.