Study objective: To determine whether the variability in the Ki67 antigen detection prevents the correct classification of breast cancer into subgroups depending on the detected Ki67. Study design: Cross-sectional retrospecitve study. Subjects and methods: The study was conducted on 50 tumor tissue samples of patients with stage T1 invasive breast cancer of no special type, diagnosed at the Osijek Cilnical Hospital Centre in 2017. Hystology slides were immunohistochemically dyed with the Ki67 antigen photographed, and a share of Ki67 positive cells was detected on at least 500 tumor cells in different regions of the tumors: in the peripheral and central regions, within the "hot spots" and beyond, as well as in three randomly selected regions. Statistical data analysis was performed using cluster analysis and by measuring the Friedman’s ANOVA and the Kendall's Coefficient of Concordance. Results: The cluster analysis has categorized 7 methods of sampling Ki67 into two groups: group 1 (tumor margins, outside hot spots, random sample 2, and tumor centre) χ2 = 18.98, p < 0,001, with a coefficient of concordance of 0.12, and group 2 (hot spots, random samples 1 and 3) χ2 = 56,28, p < 0,001, with a coefficient of concordance of 0.56. In the second phase of the statistical data analysis, the methods in which subgroups of 50 observed patients were grouped according to the first 4 and last 3 indicators were compared. These 2 methods determined 2 subgroups and then there was a statistical comparison as to whether the similar indicators were distributed equally among the patients (χ2 = 31.53, p < 0,001). In 5 patients the indicators were distributed unequally, meaning that the variation was 10 %. Conclusion: There was a 10 % variability in the data obtained from the Ki67 breast cancer antigen reading which prevents the exact classification of breast cancers into the subgroups on which the determination of patient treatment depends.