The pharmacokinetics of an antimicrobial drug in human plasma and in vitro susceptibility testing of an antimicrobial drug do not necessarily predict its efficacy in vivo. Therefore, the combined activity of an antimicrobial drug and blood-derived polymorphonuclear leukocytes (PMN) Antimicrobial drugs are traditionally tested in vitro against a series of selected clinical isolates. The results (e.g., MIC for 90% of strains tested) are compared with the levels of free drug (fraction not bound to protein) in human plasma or tissue. If these concentrations exceed the MIC during a certain time interval, it is assumed that the drug shows in vivo efficacy against the strain tested (5). However, there are many factors that may affect the reliability of these predictions (7,16,28), and important discrepancies are occasionally observed between the efficacy in vitro and the cure rate in vivo when a new antimicrobial drug is administered to patients. An example is the lack of efficacy of amoxicillinclavulanic acid in Legionnaire's disease despite low MICs (9). In contrast, azithromycin has a surprisingly good effect in chlamydial infections despite its very low levels in plasma (14). Therefore, models that take into account factors such as the unfavorable in vitro medium (e.g., low pH, low partial 02 pressure) and the characteristics of the infection to be treated, such as the intracellular persistence of the pathogen (3,12,27,29), and the bacterial persistence due to the resistance of surface-adherent microorganisms to phagocytosis and antimicrobial treatment in device-related infections (8,23) polymorphonuclear leukocytes (PMN) against a methicillinsusceptible Staphylococcus aureus were investigated, and a method which enables ex vivo study of these effects was established. For this purpose, skin cantharide blisters were provoked in human volunteers, and skin blister fluid (CBF) was sampled before and at regular intervals after a single dose of Ro 41-3399. This inflammatory exudate was incubated ex vivo with S. aureus in order to obtain time-kill curves.