It is now well demonstrated that the serum AMH level correlates tightly to the ovarian follicle content. Therefore, it presents itself as a good candidate for predicting the ovarian response to COH and for helping the decision in ART. To test this hypothesis, we used a new "second generation" enzyme immunoassay for AMH (Beckman-Coulter-Immunotech, Villepinte, France), specially designed to improve the lack of sensitivity of the previous one for the measurements of serum AMH in women. 350 women aged 21-40 yrs were selected from our IVF database for retrospective analysis. The AMH assay was performed in the serum previously collected in early follicular phase of a spontaneous cycle two to six months before a first IVF cycle. The analysis also included the results of the serum levels of FSH and inhibin B previously measured on the same serum sample as AMH, as well as the antral follicle count at ultrasonography, no follicle being larger than 9 mm. COH was performed using recFSH with cumulated doses varying from 850 to 6,075 units, under desensitization with short-acting triptorelin injections. Oocyte retrieval was performed in 323 (92%) patients, with oocyte number varying from 0 to 31 (group A, n= 65 : >15; group B, n= 223 : 5 to 14; group C, n= 35 : ≤ 4). The attempt was cancelled in 27 patients (8%) (group D) because of insufficient follicular development under COH. The mean serum AMH level was significantly different between the 4 groups (p<0.0001 by ANOVA) (group A: 43.7 ± 22.3; group B: 37.0 ± 27.3; group C: 26.6 ± 20.7 and group D: 25.3 ± 27.1 pMol/L). It correlated less tightly to the number of oocytes retrieved than the AFC (r=0.22, p<0.0005 and 0.29, p<0.0001, respectively), but stronger than FSH (-0.182, p<0.004) or inhibin B (p=0.11). FSH was the parameter the best correlated to the total number of embryos (r=-0.174, p<0.006) before the AFC (r=0.156, p<0.02) and AMH (0.125, p<0.05). The mean serum AMH level was not significantly different between cycles resulting (n=75) or not (n=275) in a clinical pregnancy (p=0.69). In order to test the predictive value of each parameter by ROC analysis, groups A and B were gathered in a single group "success" (n=288) and groups C and D in a group "failure" (n=62). The area under the ROC curve (AUC) for AMH yielded a weak value of 0.689 (95% CI: 0.608-0.77), indicating that no satisfying compromise between sensitivity and specificity could be drawn. In comparison, AUC for AFC, FSH and inhibin B was 0.624, 0.581, and 0.546, respectively Pre-treatment AMH serum level is well correlated to the main COH outcomes. However, although yielding higher AUC in ROC analysis than AFC, FSH or inhibin B, this parameter did not appear as a good predictor for poor ovarian response by itself in this retrospective series. For example, a threshold set at 20 pMol/L (2.8 ng/mL) yielded a sensitivity of 76% and a specificity of only 62%.