Clinical evidence regarding the value of 18 F-FDG PET for therapy responses assessment in breast cancer is increasing. The objective of this study is to evaluate the accuracy of 18 F-FDG PET in predicting responses to neoadjuvant therapies with meta-analysis and explore its optimal regimen for clinical use. Articles in English language relating to the accuracy of 18 F-FDG PET for this utility were retrieved. Methodological quality was assessed by QUADAS tool. Pooled estimation and subgroup analysis data were obtained by statistical analysis. Nineteen studies met the inclusion criteria and involved 920 pathologically confirmed patients in total (mean age 49.8 years, all female). Methodological quality was relatively high. To predict histopathological response in primary breast lesions by PET, the pooled sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and diagnostic odds ratio were 84% (95% CI, 78–88%), 66% (95% CI, 62–70%), 50% (95% CI, 44–55%), 91% (95% CI, 87–94%), and 11.90 (95% CI, 6.33–22.36), respectively. In regional lymph nodes, sensitivity and NPV of PET were 92% (95% CI, 83–97%) and 88% (95% CI, 76–95%), respectively. Subgroup analysis showed that performing a post-therapy 18 F-FDG PET early (after the 1st or 2nd cycle of chemotherapy) was significantly better than later (accuracy 76% vs. 65%, P  = 0.001). Furthermore, the best correlation with pathology was yielded by employing a reduction rate (RR) cutoff value of standardized uptake value between 55 and 65%. 18 F-FDG PET is useful to predict neoadjuvant therapy response in breast cancer. However, the relatively low specificity and PPV still call for caution. It is suggested to perform PET in an earlier course of therapy and use RR cutoff value between 55 and 65%, which might potentially identify non-responders early. However, further prospective studies are warranted to assess this regimen and adequately position PET in treatment management.