Although intra-operative neuromonitoring (IONM) is widely used in thyroid surgery, the validity of the received IONM signals are still unknown. Prospective collection of data forms in 29 hospitals from 8,534 patients with 15,403 nerves at risk, who underwent surgery for benign and malignant goitre disorders between August 1999 and January 2001. IONM was performed by indirect stimulation via the vagal nerve and by direct recurrent laryngeal nerve (RLN) stimulation in 12,486 cases. IONM signals were compared with early (<14 days) and late (6 months) postoperative vocal cord function findings. The transient and permanent RLN palsy rate was 2.8% and 0.7%, respectively. Monitoring of the RLN function was significantly more reliable via the indirect IONM stimulation route than via the direct IONM stimulation route (specificity P<0.05). IONM by indirect stimulation via the vagal nerve reliably excluded postoperative, permanent, vocal cord palsy (specificity 97.6%, negative predictive value 99.6%). However, a changed IONM was insufficient to predict permanent RLN palsy (sensitivity 45.9%, positive predictive value 11.6%). IONM was not associated with increased general morbidity. For intra-operative neuromonitoring, indirect stimulation of the RLN is superior to direct stimulation. An intact acoustic IONM signal is highly predictive of intact postoperative RLN function. When the IONM signal is abnormal or absent, a one-stage extensive thyroid resection should be performed only if the surgeon is absolutely convinced that the first RLN is not harmed or a total thyroidectomy is mandatory.