The optimal approach for empirical antibiotic therapy in patients with severe sepsis and septic shock remains controversial. A retrospective cohort study was conducted in the intensive care units of a university hospital. The data from 760 patients with severe sepsis or septic shock associated with Gram-negative bacteremia was analyzed. Among this cohort, 238 (31.3%) patients received inappropriate initial antimicrobial therapy (IIAT). The hospital mortality rate was statistically greater among patients receiving IIAT compared to those initially treated with an appropriate antibiotic regimen (51.7% versus 36.4%; P < 0.001). Patients treated with an empirical combination antibiotic regimen directed against Gram-negative bacteria (i.e., beta-lactam plus aminoglycoside or fluoroquinolone) were less likely to receive IIAT compared to monotherapy (22.2% versus 36.0%; P < 0.001). The addition of an aminoglycoside to a carbapenem would have increased appropriate initial therapy from 89.7 to 94.2%. Similarly, the addition of an aminoglycoside would have increased the appropriate initial therapy for cefepime (83.4 to 89.9%) and piperacillin-tazobactam (79.6 to 91.4%). Logistic regression analysis identified IIAT (adjusted odds ratio [AOR], 2.30; 95% confidence interval [CI] = 1.89 to 2.80) and increasing Apache II scores (1-point increments) (AOR, 1.11; 95% CI = 1.09 to 1.13) as independent predictors for hospital mortality. In conclusion, combination empirical antimicrobial therapy directed against Gram-negative bacteria was associated with greater initial appropriate therapy compared to monotherapy in patients with severe sepsis and septic shock. Our experience suggests that aminoglycosides offer broader coverage than fluoroquinolones as combination agents for patients with this serious infection.