We aimed to investigate the efficacy and safety of de-escalation empirical therapy for controlling infection in patients with severe aplastic anaemia (SAA) treated with antithymocyte globulin (ATG). Eighty-seven ATG-treated SAA patients who had microbiological culture-confirmed infections from 2006 to 2015 in our center were retrospectively analyzed. The efficacy of de-escalation and non-de-escalation therapy was compared. Among all 87 patients, 63 patients were treated with de-escalation therapy and 24 patients with non-de-escalation therapy. More patients showed response to anti-infection treatment in de-escalation group than in non-de-escalation group both on day 7 (60.32% vs. 25.00%, P = 0.003) and on day 30 (79.37% vs. 58.33%, P = 0.047) since the initial antimicrobial therapy. On day 30, more patients had increased absolute neutrophil count in de-escalation group compared with non-de-escalation group (76.19% vs. 45.83%, P = 0.007), and de-escalation group had lower morality rate (17.46% vs. 37.50%, P = 0.047) and better survival outcome (P = 0.023) on day 90. Twenty-three patients in de-escalation group and 5 patients in non-escalation group received granulocyte transfusions. Granulocyte transfusions helped to control infections in both de-escalation group (P = 0.027) and non-de-escalation group (P = 0.042) on day 7, but did not improve survival on day 90. We concluded that de-escalation antibiotics improved survival in SAA patients after ATG treatment. Early administration of broad-spectrum antibiotics pending microbiological cultures combined with a commitment to change to narrow-spectrum antibiotics should be recommended for controlling infections in SAA patients treated with ATG. Granulocyte transfusions might be an adjunctive therapy in controlling infections.