Background Neutropenia after chemotherapy can frequently lead to a life-threatening infection. Unexpected episodes of neutropenia can occur after the first cycle of chemotherapy. We sought to characterise predictive risk factors for severe neutropenia in patients with breast cancer after the first cycle of chemotherapy. Methods We prospectively collected data for patients with breast cancer who received the doxorubicin and cyclophosphamide (AC) regimen for any stage at the oncology unit during January to December 2011. Patients who received primary CSF prophylaxis were not included. The correlation between patient’s demography, breast cancer history, blood chemistry, and occurrence of severe neutropenia and febrile neutropenia (FN) were analysed. Findings Seventy-five patients with breast cancer were included in this study. Their mean age was 49 years (SD±9). Most patients (98.9%) had a good performance status. None had received previous chemotherapy or radiotherapy. There were a few patients with comorbidities, including 1% with type 2 diabetes mellitus, and 19% with hypertension. 4% of the breast cancers were metastatic to the lung and bone. The mean body surface area was 1.59m2. All patients had normal baseline white blood cell counts. After the first cycle of AC, 84% and 60% of patients developed grade 3 and 4 neutropenia, respectively, which turned into FN in 20% of patients. Analyses of predictive factors showed no statistically significant correlation between grade 4 neutropenia and age greater than 60 years (RR 0.95, 95% confidence interval (CI) 0.49–2.06), bovine serum albumin (BSA) less than 1.45m2 (RR 1.10, 95% CI 0.73–1.67), underweight for body-mass index less than 18.5kg/m2 (RR 1.0, 95% CI 0.48–2.1), dietary protein index less than 0.5g/kg/day (RR 1.05, 95% CI 0.50–2.22), and non-metastatic disease (RR 0.896, 95% CI 0.39–2.04). Interpretation Simple clinical factors cannot be used to reliably predict the risk of FN in patients with breast cancer during the first cycle of chemotherapy with AC. For development of future predictive models, the complex relation within datasets should be taken into account such as novel biomarkers or genetic profiles. The authors declared no conflicts of interest.