(Doll, 1990;Mann, 1990). The global evidence on the influence of OC use on the breast in all age groups is largely reassuring (Prentice & Thomas, 1987;Thomas, 1988;Doll, 1990;Mann, 1990;Olsson, 1989;Delgado-Rodriguez et al., 1991;La Vecchia, 1992), and in a formal overview based on 16 studies and over 12,000 cases, an overall relative risk (RR) for over use of 1.0 (95% confidence interval (CI) = 0.9-1.1) was found (Thomas, 1988).A positive association between OC and breast cancer has been reported in several subgroups of women, but not always consistently. There is convincing evidence that long-term pill use increases the risk of breast cancer in women before age 35 or 45 (Lubin et al., 1982;Meirik et al., 1986;Miller et al., 1989;Olsson et al., 1989;Peto, 1989; UK National CaseControl Study Group, 1989;Delgado-Rodriguez et al., 1991;Rushton & Jones, 1992;Ursin et al., 1992), also in the absence of any evidence of an association in older women (Romieu et al., 1990;McPherson et al., 1987;Kay & Hannaford, 1988;Rosenberg et al., 1984;Weinstein et al., 1991). Increased risks have also been reported for use before first term pregnancy (Pike et al., 1981;Meirik et al., 1986;McPherson et al., 1987), but these results are not consistent (Vessey et al., 1982;Stadel et al., 1985;Paul et al., 1986).OC use may become a risk factor only after a long 'latent period', so that now we may be observing only the start of a pill-induced breast cancer epidemic (McPherson et al., 1987). However, this is not borne out by several studies which found no relationship between time since first use and subsequent breast cancer risk (Brinton et al., 1982; Schesselman et al., 1988; Vessey et al., 1989;Ewertz, 1992).To provide further information on this issue, we report here the final update of a case-control study conducted in Northern Italy (La Vecchia et al., 1986a Mantel and Haenszel (1959); for multiple levels of exposure, the significance of the linear trend in risk was assessed by the Mantel test. Unconditional multiple logistic regression, fitted by the method of maximum likelihood, was used to allow for several possible confounding factors (Breslow & Day, 1980).