STUDY QUESTION What is the clinical value of anti-Müllerian hormone (AMH) for the prediction of high or low ovarian response in controlled ovarian stimulation for IVF using GnRH antagonist treatment? SUMMARY ANSWER AMH as a single test has substantial accuracy for ovarian response prediction in GnRH antagonist treatment for IVF, with a higher accuracy for predicting a high response than for low response. WHAT IS KNOWN ALREADY The role of AMH and other patient characteristics in ovarian response prediction has been studied extensively in long GnRH agonist protocols; however, little information is available regarding the clinical value in GnRH antagonists. STUDY DESIGN, SIZE, DURATION This is an observational (retrospective) substudy as part of an ongoing cohort study. A total of 487 patients scheduled for IVF/ICSI between 2006 and 2011 were included in the study. PARTICIPANTS/MATERIALS, SETTING, METHODS Patients with a regular cycle who underwent their first IVF/ICSI cycle with GnRH antagonist treatment while receiving a starting dose of 150 or 225 IU recombinant FSH were included in the study. Patients were divided into three subgroups according to the following ovarian response categories: high (>15 oocytes or cycle cancellation), normal (4-15 oocytes) and low (<4 oocytes or cycle cancellation). Serum samples collected prior to IVF treatment were used to determine serum AMH levels. MAIN RESULTS AND THE ROLE OF CHANCE According to the predefined ovarian response categories, 58 patients were classified as high, 326 as normal and 101 as low responders, and the ongoing pregnancy rates did not differ among groups (19.0, 22.1 and 16.8%, respectively, P = 0.9). For the prediction of high response, AMH had an area under the receiver-operating characteristic curve (AUC) of 0.87. Both female age and BMI had lower accuracy (AUC 0.66 and 0.58, respectively). For low response prediction, again AMH had a better accuracy (AUC 0.79) than female age and BMI (AUC 0.59 and 0.56, respectively). In a multivariate model, including the factors age, AMH, BMI, smoking, type and duration of subfertility, only BMI added some predictive value to AMH for both high and low response prediction. Clinical test characteristics demonstrated that using a specificity of ∼90%, the detection rate of AMH for high and low response, corresponding with a test cut-off of 4.5 and 0.8 μg/l, was ∼60 and ∼45%, respectively. LIMITATIONS, REASONS FOR CAUTION The impact of the antral follicle count (AFC) on ovarian response prediction in GnRH antagonists was not assessed; however, previously studies demonstrated that for GnRH antagonists, AMH has a better accuracy than AFC. WIDER IMPLICATIONS OF THE FINDINGS The current study demonstrates that AMH is an adequate predictor for both high and low response in GnRH antagonist cycles, showing a similar accuracy to GnRH agonists, as reported previously. The optimization and individualization of GnRH antagonist protocols may be improved by using an AMH-tailored approach. STUDY FUNDING/COMPETING INTEREST(S) This study was funded by the Academic Institutional Resources of the Department of Reproductive Medicine of the UMC Utrecht. O.H., M.J.C.E, E.W.G.L and H.L.T. have nothing to declare. N.S.M. has received fees and/or grant support from the following companies (in alphabetic order): Anecova, Ferring, Informa, Merck Serono and MSD. B.C.J.M.F. has received fees and/or grant support from the following companies (in alphabetic order); Childhealth, CVON, Ferring, Ova-Science, PregLem, Roche and Watson laboratories. F.J.B. has received fees and/or grant support from the following companies (in alphabetic order); Merck Serono and MSD. TRIAL REGISTRATION NUMBER www.clinicaltrials.gov, Protocol ID 13-109.