Background: Chronic obstructive pulmonary disease (COPD) phenotyping using stable-state blood eosinophil level was shown to have prognostic implication in terms of exacerbation risk. However, using a single cut-off of blood eosinophil level to predict clinical outcome has been challenged. There have been suggestions that variability of blood eosinophil count at stable-state could provide additional information on exacerbation risk. Methods: A retrospective cohort study was conducted in a major regional hospital and a tertiary respiratory referral centre in Hong Kong, including 275 Chinese patients with COPD, to investigate the possible role of variability of blood eosinophil count at stable-state to predict COPD exacerbation risk in one year. Results: Higher variability of baseline eosinophil count, which is defined as the difference of the minimal and maximal eosinophil count at stable-state, was associated with increased risk of COPD exacerbation in the follow-up period with adjusted OR (aOR) of 1.001 (95% CI = 1.000-1.003, p-value = 0.050) for 1 unit (cells/µL) increase in variability of baseline eosinophil count, aOR of 1.72 (95% CI = 1.00-3.58, p-value = 0.050) for 1 SD increase in variability of baseline eosinophil count and aOR of 1.06 (95% CI = 1.00-1.13) for 50 cells/µL increase in variability of baseline eosinophil count. The AUC by ROC analysis was 0.862 (95% CI = 0.817-0.907, p-value < 0.001). The cut-off for variability of baseline eosinophil count identified was 50 cells/µL, with sensitivity of 82.9% and specificity of 79.3%. Similar findings were also shown in the subgroup with stable-state baseline eosinophil count below 300 cells/µL. Conclusion: Variability of baseline eosinophil count at stable-state might predict the exacerbation risk of COPD, exclusively among patients with baseline eosinophil count below 300 cells/µL. The cut-off value for variability was 50 cells/µValidation of the study findings in large scale prospective study would be meaningful.