The aim of this retrospective study was to create guidelines for therapy of severe sepsis in surgical intensive care unit (ICU) for unknown causative agent based on antimicrobial susceptibility of causative bacteria. Seventy-four patients with severe sepsis from surgical ICU in 2003.-2005. were included in study. Their clinical and microbiological data were analyzed from the medical records. Antimicrobial susceptibility of the strains isolated from the blood-culture was tested by disk diffusion method according to CLSI (Clinical Laboratory Standard Institution). APACHE II score was used to predict the severity of illness. Statistical significance difference between results was tested by Mann-Whitney test and chi2 test. Important problem remained type of sepsis: mono-agent sepsis presented less therapeutic problem than sepsis caused with two or more agents (mixed sepsis). Methicillin-resistant Staphylococcus aureus (MRSA), Pseudomonas aeruginosa and Acinetobacter baumannii were predominant causative agents in both type of sepsis. There was remarkable increase of A. baumannii prevalence in 2005 compared to 2004 and to 2003. There was also decrease of MRSA prevalence in 2004 and 2005 compared to 2003. P. aeruginosa were the predominant causative agents in 2004. MRSA displayed good susceptibility to vancomycin and linezolide, whereas P. aeruginosa showed excellent susceptibility to ceftazidime and carbapenems. A. baumannii, third predominant causative agent, exhibited excellent susceptibility to ampicillin+ sulbactam and carbapenems. The recommended therapy is empirical and should cover all important pathogens.