We aim to investigate the diagnostic capability of single-voxel proton MR spectroscopy (MRS) for benign/malignant discrimination of focal breast lesions with a meta-analysis. The meta-analysis included a total of 750 malignant breast lesions and 419 benign breast lesions from eighteen studies. The pooled sensitivity and specificity of MRS were 0.71 (95 % CI 0.68-0.74) and 0.85 (95 % CI 0.81-0.88), respectively. The positive likelihood ratio and negative LR were 4.11 (95 % CI 3.11-5.43) and 0.25 (95 % CI 0.17-0.36), respectively. The P value for chi (2) heterogeneity for all pooled estimates was < 0.05. From the fitted summary receiver operating characteristics curve, AUC was 0.89 and Q* was 0.84. Asymmetrical in funnel plots indicated there may be publication bias (t = 2.85, P = 0.012). The meta-regression analysis indicated that neither threshold effect nor evaluated covariates that include strength of field, scanning technique (PRESS or STEAM), repetition time, NSA, and pre- or post-contrast agent were the sources of heterogeneity (all P value > 0.05). Single-voxel proton MRS was useful for differentiation between malignant and benign breast lesions. However, pooled diagnostic measures might be overestimated. The standardization of the acquisition protocol for MRS across the multicenter trials is recommended.