The objectives of the present study were to evaluate in vitro the production of antagonistic compounds against Gardnerella vaginalis by Lactobacillus strains isolated from women with or without bacterial vaginosis (BV), and to select one of the better Lactobacillus producers of such a substance to be tested in vivo using a gnotobiotic animal model challenged with one of the more sensitive G. vaginalis isolates. A total of 24 isolates from women with and without BV were identified as G. vaginalis. A higher frequency (P<0.05) of this bacterium was observed in women with BV (56.7%) when compared to healthy women (17.6%). A total of 86 strains of Lactobacillus were obtained from healthy women and women with BV. Lactobacillus strains were more frequently present (P<0.05) in healthy women (97.5%) than in women with BV (76.7%). Lactobacillus crispatus was the predominating strain in both healthy women and women with BV. Lactobacillus jensenii, Lactobacillus johnsonii, Lactobacillus gasseri and Lactobacillus vaginalis were isolated with an intermediate frequency in the two groups. In vitro antagonism assays were performed using as indicators 17 reference strains and the G. vaginalis strains isolated from women with BV and from healthy women. Lactobacillus isolated from healthy women showed the higher antagonistic activity against all the indicator strains when compared with isolates from women with BV. Concerning the indicator strains, G. vaginalis found in women with BV was more resistant to the antagonism, particularly when Lactobacillus isolates from women with BV were used as producer strains. A high vaginal population level of G. vaginalis was obtained by intravaginal inoculation of germ-free mice, and this colonization was accompanied by vaginal histopathological lesions. A tenfold decrease in vaginal population level of G. vaginalis and a reduction of histological lesions were observed when the pathogenic challenge was performed in mice previously monoassociated with an L. johnsonii strain. Concluding, results of the present study suggest that progression of G. vaginalis-associated BV depends in part on a simultaneous presence of Lactobacillus populations with a low antagonistic capacity and of a G. vaginalis strain with a high resistance to this antagonism. The results could also explain why G. vaginalis is frequently found in the vaginal ecosystem of healthy women.