In-vitro and animal model data indicate that the time beta-lactam serum concentrations remain above the MIC is an important determinant of the organism kill at the primary infection site. Similarly, for aminoglycosides, area under the curve and peak concentrations have been linked to organism kill and suppression of resistance. It is only in clinical patients that these data can be validated as to their significance. For beta-lactams, little clinical data exist regarding these concepts. However, Bodey & colleagues have shown that profoundly, persistently neutropenic cancer patients fared better when one of their beta-lactams was administered continuously. Our group was able to correctly predict outcome in 9/10 patients bacteremic with a Gram-negative bacillus when receiving a single beta-lactam on the basis of the time free drug concentrations remained above the MIC. Schentag et al studied patients with lower respiratory tract infection treated with cefmenoxime alone and found a significant relationship between time greater than MIC & time to clearance of the pathogen from cultures of the tracheobronchial tree. These data would seem to validate the predictive nature of the findings from in-vitro & animal model studies. With aminoglycosides, Moore, Smith & Lietman were able to demonstrate a highly significant correlation between outcome and the maximal peak concentrations to MIC ratio achieved for patients with single organism Gram-negative rod infections. This is somewhat at variance with some animal models, but as the studies were performed with a fixed dosing interval, the outcome is not surprising. Little has been done with combinations of these agents in patients. Barriere & colleagues have proposed the AUC of the reciprocal serum bactericidal activity curve as a way to integrate the activity of combinations. We have developed a method employing logistic regression analysis to integrate the activity from the administration of multiple agents. The integration of this approach with each drug's pharmacokinetics allows the generation of a plot of the probability of the blood remaining sterile over a steady state dosing interval. This approach has been preliminarily tested in 6 individuals with excellent concordance between outcome and prediction. Development of data in-vitro and in animal models with validation in patients will hopefully provide the impetus to optimize therapy, and thence, outcome for the most seriously ill individuals.