The predictive value of in vitro studies of antibiotic interaction for clinical drug interactions is unclear. Five clinical isolates (two Klebsiella, two Pseudomonas aeruginosa, and one Serratia marcescens) were evaluated by the time-kill curve method for in vitro synergy between amikacin and imipenem. When we used the stringent definition of synergy of Hallander et al., no synergy was present for any study strain; however, when we used a more-conventional definition of synergy, these drugs interacted synergistically against all study strains. The results of these in vitro studies were correlated with in vivo interactions by using neutropenic infant rats injected ip with study organisms and given various treatment regimens. For 80% of the study strains, treatment of rats with amikacin and imipenem resulted in significantly greater survival than did therapy with either drug alone or than could be predicted by addition of survival rates achieved with either agent alone (P less than .005). In vitro studies predicted this in vivo synergy in 80% of the cases when the more-conventional definition of synergy was used, whereas they were not predictive when the more-stringent definition of synergy was used. This rat model of neutropenia and gram-negative sepsis may provide more insight into in vivo drug interactions than do current methods.