There are many clinical studies comparing antibiotic treatments, but the majority are too small to have sufficient power to be reasonably certain of detecting statistically a moderate treatment effect. For example, few of the 19 studies published on imipenem-cilastatin for empirically treating febrile neutropenic patients were able to show any significant treatment effect in either direction when compared with alternative regimens. We therefore undertook a meta-analysis of these studies and made 21 pairwise comparisons of a control regimen with imipenem-cilastatin. Eleven comparisons were made between imipenem-cilastatin and a beta-lactam-aminoglycoside combination, and a further 10 between the carbapenem and a beta-lactam regimen either alone or combined with a glycopeptide or other beta-lactam antibiotic. These two data sets were analysed separately. Imipenem-cilastatin demonstrated a beneficial treatment effect over that achieved by aminoglycoside containing control regimens, yielding a typical odds ratio (OR) of 0.77 (95% CI 0.61 to 0.98). A beneficial treatment effect for the carbapenem regimen was also shown against regimens that did not include an aminoglycoside, with the typical OR being 0.67 (95% CI 0.54 to 0.84). Although there was clinical heterogeneity between studies, the treatment effect itself was relatively homogenous. These results show meta-analysis to be a useful aid for interpreting the data from clinical studies that are intrinsically too small to provide conclusive results.