Objectives: To analyse the diagnostic performance of eosinopenia, alone or combined with polymorphonuclear neutrophils (PMN) and/or lymphocytes, as a marker of active COVID-19 in patients hospitalized for suspicion of SARS-CoV-2 infection. Methods: A prospective observational study including patients hospitalized for suspicion of COVID-19 in a COVID unit was performed from 20th March to 5th April 2020, in Perpignan, France. Patients for which there was a doubt upon diagnosis, who were recently under oral corticosteroids, had myeloid malignancy or human immunodeficient virus infection were excluded. SARS-CoV-2 detection was performed using an RT-PCR assay, from nasopharyngeal swab specimens. Complete blood count were performed for all patients. Results: One-hundred and twenty-one patient were included: 57 patients were diagnosed with COVID-19, 64 patients were not. Eosinophil count was lower in the COVID-19 group (median: 0/μL versus 70/μL, p < 0.0001). To diagnose COVID-19, eosinopenia had a sensitivity of 89.5% and a specificity of 78.1% while lymphopenia's were 73.7% and 62.5% respectively. Using area under curve (AUC) of receiving operating characteristics (ROC) curves, eosinophil's optimal cut-off level was 10/μL, sensitivity and specificity were 86%, and 79.7% respectively. Regarding the eosinophil/PMN ratio, the optimal cut-off level was 3.344, sensitivity and specificity were 87.7% and 73.4% respectively. The AUC of lymphocyte/PMN ratio was significantly lower than eosinophil/PMN ratio's (0.621 versus 0.846, p = 0.0003). Conclusion: Eosinopenia - <10/μL - and eosinophil/PMN ratio are useful, low-cost, reproducible tools to help diagnose COVID-19, during an epidemic period, in a population of hospitalized patients admitted for suspicion of COVID-19.