The benchmark dose (BD) approach has been applied to foetal data from four gavage segment II studies (rat studies 1 and 2, rabbit study, hamster study) on the teratogenic benzimidazole carbendazim. Nineteen parameters were assessed using the log-normal model as a practical tool to derive BDs; good model fitting was observed for all except two parameters. Data were evaluated on a 'per-implant/foetus' basis; BDs were derived from response rate increases of 1, 5, and 10%. The values were compared to the lowest-observed-adverse-effect levels (LOAELs) and no-observed-adverse effect levels (NOAELs) obtained by Fisher's exact test on a 'per-implant/foetus' basis. Frank effects observed only at the top dose and/or small sample size tended to increase the 95% confidence limits and this influenced the determination of BD. Generally, the BD approach provided slightly more conservative estimates than NOAEL; overall, BD01 and BD05 were similar to NOAEL, or even lower for several parameters. The LOAEL in most cases was similar to BD10. Reference doses obtained by dividing BD01 by a 10 or 100 uncertainty factor, corresponded to residual risks of 10(-5) or below. For two critical parameters (hydrocephalus in rat study 1 and resorption rate in the rabbit study) a NOAEL could not be found, whereas a BD was always determined.