Mild forms of hypothyroidism--subclinical hypothyroidism--have recently been discussed as being a risk factor for the development of overt thyroid dysfunction and for a number of clinical disorders. The diagnosis critically depends on the definition of the upper normal limit of serum TSH as, by definition, free thyroxine serum concentrations are normal. Cut-off levels of 4-5 mU TSH/l have been conventionally used to diagnose an elevated TSH serum concentration. Recent data from large population studies have suggested a much lower TSH cut-off with an upper limit of 2-2.5 mU/l but application of strict criteria for inclusion of subjects from the general population studies aiming at assessing TSH reference intervals (no personal or family history of thyroid disease, no thyroid antibodies and a normal thyroid on ultrasonography) did not result in an unequivocal upper limit of normal TSH at 2.0-2.5 mU/l. When summarizing the available evidence for lowered upper TSH cut-off values and their potential therapeutic implications there is presently insufficient justification to lower the upper normal limit of TSH and, for practical purposes, it is still recommended to maintain the TSH reference interval of 0.4-4.0 mU/l. Classifying subjects with a TSH value between 2 and 4 mU/l as abnormal, as well as intervening with thyroxine treatment in such subjects, is probably doing more harm than good.