Background: It is now well established that AMH can indicate the extent of chemotherapy-induced damage on the ovarian reserve. Very low AMH levels in women after chemotherapy for early breast cancer (eBC) may indicate treatment-induced premature ovarian failure (POI), but the evidence to implement its routine clinical use remains limited. Methods: AMH was measured (Roche cobas autoanalyser) in 197 women aged 40-45 years at diagnosis treated with anthracycline/cyclophosphamide/taxane based chemotherapy for eBC, with no previous endocrine therapy and no subsequent aromatase inhibitor exposure or ovarian suppression. 76% were ER positive and received tamoxifen. The primary objective was to assess the relationships between AMH levels at 6 months after chemotherapy (AMH6) with subsequent ovarian function at 30 months. This included accuracy of prediction of loss of ovarian function at 30 months (E30) after chemotherapy by AMH levels at 6 months (AMH6) and the value of other endocrine/clinical factors pretreatment and at 6 months. Multivariate analysis was performed with data were split into 70% training and 30% testing sets with results reported for model performance on the test set. Results: AMH fell from a median of 0.62 (IQR 0.21-1.31) ng/ml pretreatment to become undetectable ( < 0.01ng/ml) in 137 (70%) of women at 6 months. AMH showed very little recovery, being undetectable in 115 women at 18 months and 119 at 30 months. In those with detectable AMH at 30 months, it remained very low, median 0.07ng/ml, max 0.82ng/ml but nevertheless it reflected ovarian function. Thus median estradiol at 30 months (E30) was 50 pmol/l (IQR 34-68) in women with undetectable AMH at that time, vs 313 pmol/l (IQR 102-1052) in the 80 women with detectable AMH. AMH at 6 months was predictive of ovarian function at 30 months, as E30 was 56 pmol/l (IQR 40-104) in women with undetectable AMH6 vs 258pmol/l (IQR 69-780; p < 0.0001) in women with detectable AMH6, and the positive predictive value (PPV) of AMH being undetectable at 6 months and remaining so at 30 months was 77%. Random forest analysis identified BMI as a non-endocrine predictive factor of E30. Using a stringent cutoff of ≤50pmol/l (13.6 pg/ml) and adding 1:9 weighting to improve PPV gave a test validated PPV of 79.0% while maintaining AUC at 83.7% for prediction of E30. Other classifiers performed less well than random forests. Conclusions: Measurement of AMH at 6 months after chemotherapy has predictive value for later ovarian function, and combining this with other endocrine and baseline data improves discriminative ability. This approach indicates the potential role of multivariate analysis based on AMH measured shortly after chemotherapy to helping optimise subsequent endocrine therapy.