Rational: Severe sepsis is one of the leading cause of mortality in intensive care unit patients. Early initiation of an appropriate empirical antimicrobial therapy is associated with improved outcomes. In order to avoid an increase of selection pressure and the emergence of multidrug resistant pathogens, guidelines recommend to streamline the antimicrobial therapy after the identification of the pathogen responsible for infection. This strategy has been evaluated in several observational studies. However, at the bedside, few randomized clinical trials tested this strategy prospectively.

Method: the investigators conduct a randomized clinical trial comparing a strategy based on de-escalation (streamlining of the empirical antimicrobial therapy) and a conservative strategy (continuation of the empirical antimicrobial therapy). The investigators first aim was to show that a strategy based on de-escalation is not inferior to a conservative strategy in terms of intensive care unit length of stay. Secondary aims are to compare the rate of mortality rate, the emergence of multidrug resistant pathogens, and the feasibility of de-escalation. The study is performed in nine intensive care units from four institutions, and 120 patients are required to validate the investigators hypothesis. New technologies for the rapid diagnosis of severe infections are investigated in an ancillary study.

Inclusion Criteria:

• Major Subject;
• Subject having a sepsis engraves(burns) defined according to the following criteria during the initiation of the probability antibiotic treatment:
• Criteria of SIRS [ 14 ],
• And a suspected infection,
• And a failure of organ: low blood pressure, respiratory failure, coma, hepatic insufficiency, renal insufficiency, thrombopénia, spontaneous extension of the TCA,
• Subject for which an antibiotic treatment was begun within 6 hours following the diagnosis of sepsis engraves(burns);
• Subject for which a taking with microbiological aim was made within 48 hours following the diagnosis of sepsis

Exclusion Criteria:

• Minor Subject, pregnant or breast-feeding woman;
• Neutropénia (PN < 1000 / mm3);
• Absence of identification of a microorganism in the microbiological examinations;
• Absence of Social Security;
• Subject deprived of freedom or under guardianship;
• Subject for which the lit(enlightened) consent is not collected(taken in) (itself and/or reliable person).