To assess the effect of five different surgical treatments for glue ear (secretory otitis media) on improvement in hearing and, assuming one or more treatments to be effective, to identify the appropriate indications for surgery. Randomised controlled trial of children receiving (a) adenoidectomy, bilateral myringotomy, and insertion of a unilateral grommet; (b) adenoidectomy, unilateral myringotomy, and insertion of a unilateral grommet; (c) bilateral myringotomy and insertion of a unilateral grommet; and (d) unilateral myringotomy and insertion of a grommet. Children were followed up at seven weeks, six months, 12 months, and 24 months by symptom history and clinical investigations. Otolaryngology department in an urban hospital. 149 Children aged 4-9 years who were admitted for surgery for glue ear and who had no history of previous operations on tonsils, adenoids, or ears and no evidence of sensorineural deafness. Inadequate follow up information on levels of hearing and on middle ear function was obtained from 22. Mean hearing loss (dB) of the three worst heard frequencies between 250 and 4000 Hz, results of impedance tympanometry, and parental views on their child's progress. In the 127 children for whom adequate information was available ears in which a grommet had been inserted performed better in the short term (for at least six months) than those in which no grommet had been inserted, irrespective of any accompanying procedure. Most of the benefit had disappeared by 12 months. Adenoidectomy produced a slight improvement that was not significant, though was sustained for at least two years. The ears of children who had had an adenoidectomy with myringotomy and grommet insertion, however, continued to improve so that two years after surgery about 50% had abnormal tympanometry compared with 83% of those who had had only myringotomy and grommet insertion, and 93% of the group that had had no treatment. Logistic regression analyses identified preoperative hearing level as the single best predictor of good outcome from surgery. Other variables contributed little additional predictive power. If the principal objective of surgery for glue ear is to restore hearing then our study shows that insertion of grommets is the treatment of choice. The addition of an adenoidectomy will increase the likelihood of restoration of normal function of the middle ear but will not improve hearing. When deciding appropriate indications for surgery, a balance has to be made between performing unnecessary operations and failing to treat patients who might benefit from surgical intervention. Preoperative audiometry scores might be the best predictor in helping to make this decision.