OBJECTIVE To investigate to what extent thyroid remnant ablation and withdrawal from thyroxine are required to achieve sufficient accuracy of serum thyroglobulin (Tg) measurements as an indicator of tumour recurrence in the follow-up of patients with differentiated thyroid carcinoma. METHODS We conducted a meta-analysis of the literature from 1975 to 2003 on serum Tg measurements in the follow-up of differentiated thyroid carcinoma. In a computer-based search, we initially found 915 articles that were finally narrowed down to 120. These 120 papers were subjected to strict in/and exclusion criteria, leaving 46 articles (totalling 9094 patients). Data from these articles were extracted in a structured fashion and were grouped according to initial therapy, TSH status, Tg assay method and definition of a 'gold standard'. Original 2 x 2 tables were pooled by summary receiver operating characteristic curve analysis (sROCa), best estimates of sensitivity and specificity being obtained by the combination of sROCa and Mantel-Haenszel odds ratios. RESULTS Despite considerable differences between series in laboratory and clinical methodology, we consistently found higher specificity for Tg measurements after thyroid remnant ablation than after surgery alone. Highest pooled sensitivity 0.961 +/- 0.013 (SE) was found for immunometric assay (IMA) after thyroid remnant ablation and thyroid hormone withdrawal, at a specificity of 0.947 +/- 0.007. Pooled sensitivity decreased significantly if ablated patients were tested while on thyroid hormone (0.778 +/- 0.023, at a specificity of 0.977 +/- 0.005). Significantly decreased pooled specificity was found in patients who did not undergo remnant ablation (sensitivity 0.972 +/- 0.023, at a specificity of 0.759 +/- 0.028). If recombinant human TSH (rhTSH) stimulation was used as a substitute for thyroxine withdrawal, sensitivity remained high (0.925 +/- 0.018) while specificity decreased to 0.880 +/- 0.013. In all analyses, specificity of Tg would decrease when unspecified activity in the thyroid region at scintigraphy was considered benign, whereas sensitivity decreased when such activity was considered malignant. CONCLUSIONS This study confirms that the best accuracy of Tg-guided follow-up in patients treated for differentiated thyroid carcinoma is obtained if treatment includes remnant ablation, and Tg testing is performed while off thyroxine.