The aim of this study was to evaluate the diagnostic performance of the use of total choline signal-to-noise ratio (tCho SNR) criteria in MRS studies for benign/malignant discrimination of focal breast lesions. We conducted (1) a meta-analysis based on 10 studies including 480 malignant breast lesions and 312 benign breast lesions and (2) a subgroup meta-analysis of tCho SNR ≥ 2 as cutoff for malignancy based on 7 studies including 371 malignant breast lesions and 239 benign breast lesions. (1) The pooled sensitivity and specificity of proton MRS with tCho SNR were 0.74 (95 % CI 0.69-0.77) and 0.76 (95 % CI 0.71-0.81), respectively. The PLR and NLR were 3.67 (95 % CI 2.30-5.83) and 0.25 (95 % CI 0.14-0.42), respectively. From the fitted SROC, the AUC and Q* index were 0.89 and 0.82. Publication bias was present (t = 2.46, P = 0.039). (2) Meta-regression analysis suggested that neither threshold effect nor evaluated covariates including strength of field, pulse sequence, TR and TE were sources of heterogeneity (all P value >0.05). (3) Subgroup meta-analysis: The pooled sensitivity and specificity were 0.79 and 0.72, respectively. The PLR and NLR were 3.49 and 0.20, respectively. The AUC and Q* index were 0.92 and 0.85. The use of tCho SNR criteria in MRS studies was helpful for differentiation between malignant and benign breast lesions. However, pooled diagnostic measures might be overestimated due to publication bias. A tCho SNR ≥ 2 as cutoff for malignancy resulted in higher diagnostic accuracy.