We compared metronidazole 0.75% gel (containing 37.5 mg metronidazole per dose) with ovules containing metronidazole 500 mg and nystatin 100,000 U, for intravaginal treatment of bacterial vaginosis (BV). In a single-blinded trial, symptomatic women with BV by both Amsel and Nugent criteria were randomly assigned to gel or ovules, once nightly for 5 nights, and asked to return 3 times after treatment. Analyses were intent-to-treat. Of 151 women with BV by both criteria at enrollment, 138 (91%) returned at least once. Product limit estimates for persistence or recurrence of BV at 14, 42, and 104 days were 20% (95% CI 10%-29%), 38% (95% CI 25%-48%), and 52% (95% CI 37%-63%) after gel treatment, and 4% (95% CI 0%-9%), 17% (95% CI 7%-26%), and 33% (95% CI 21%-46%) after ovule treatment (P =. 01). Among women without BV at first follow-up, subsequent intercourse without condoms independently predicted subsequent recurrence (P ≤. 01). Metronidazole/nystatin ovules were significantly more effective than metronidazole gel. Unprotected sex predicted recurrence after initial improvement. © 2004 Elsevier Inc. All rights reserved.