Purpose Metformin is associated with reduced breast cancer risk in observational studies in patients with diabetes, but clinical evidence for antitumor activity is unclear. The change in Ki-67 between pretreatment biopsy and post-treatment surgical specimen has prognostic value and may predict antitumor activity in breast cancer. Patients and Methods After tumor biopsy, we randomly allocated 200 nondiabetic women with operable breast cancer to either metformin 850 mg/twice per day (n = 100) or placebo (n = 100). The primary outcome measure was the difference between arms in Ki-67 after 4 weeks adjusted for baseline values. Results Overall, the metformin effect on Ki-67 change relative to placebo was not statistically significant, with a mean proportional increase of 4.0% (95% CI, −5.6% to 14.4%) 4 weeks apart. However, there was a different drug effect depending on insulin resistance (homeostasis model assessment [HOMA] index > 2.8, fasting glucose [mmol/L] × insulin [mU/L]/22.5; P_interaction = .045), with a nonsignificant mean proportional decrease in Ki-67 of 10.5% (95% CI, −26.1% to 8.4%) in women with HOMA more than 2.8 and a nonsignificant increase of 11.1% (95% CI, −0.6% to 24.2%) with HOMA less than or equal to 2.8. A different effect of metformin according to HOMA index was noted also in luminal B tumors (P_interaction = .05). Similar trends to drug effect modifications were observed according to body mass index (P = .143), waist/hip girth-ratio (P = .058), moderate alcohol consumption (P = .005), and C-reactive protein (P = .080). Conclusion Metformin before surgery did not significantly affect Ki-67 overall, but showed significantly different effects according to insulin resistance, particularly in luminal B tumors. Our findings warrant further studies of metformin in breast cancer with careful consideration to the metabolic characteristics of the study population.