To investigate the clinical outcome of cytomegalovirus (CMV) infection in febrile hospitalized patients with autoimmune diseases, mostly systemic lupus erythematosus (SLE). Fifty-four febrile patients were analyzed retrospectively. Half were diagnosed as CMV infection, by positive CMV pp65 antigenemia assay. Clinical and laboratory data between two groups were compared. Correlation between laboratory data and SELENA-SLEDAI scores/mortality were analyzed in the CMV infection group. Receiver operating characteristic analysis was performed to determine the cutoff points of different parameters for predicting mortality or morbidity. The CMV infection group received a higher corticosteroid dosage (mean 26.3 mg/day) and a higher percentage of azathioprine use before admission than the non-CMV infection group. In the former, the deceased subgroup had a significantly higher number of infected leukocytes for CMV (shortened as CMV counts, P = 0.013), more cases of bacterial infection (P = 0.090), and a higher SLE disease activity index score (P = 0.072) than the alive subgroup. The CMV infection group had lower lymphocyte count and more positive bacterial infection than the non-CMV infection group did (P = 0.013 and P = 0.027, respectively). A level of 25 CMV particles/5 × 105 polymorphonuclear neutrophils (PMN) was the best cutoff point for predicting CMV-associated mortality, with a sensitivity of 75.0% and specificity of 72.2%. Moderate dose (30 mg/day) of prednisolone or azathioprine use predisposes patients with autoimmune diseases to CMV infection with concurrent bacterial infection. In particular, peak CMV counts at 25/5 × 105 PMN or low lymphocyte counts predict mortality or morbidity, respectively.