OBJECTIVE--To investigate the predictors of first-round attendance for breast screening in an inner city area. DESIGN--Prospective design in which women were interviewed or completed a postal questionnaire before being sent their invitation for breast screening. Sociodemographic factors, health behaviours, and attitudes, beliefs, and intentions were used as predictors of subsequent attendance. A randomised control group was included to assess the effect of being interviewed on attendance. SETTING--Three neighbouring health districts in inner south east London. PARTICIPANTS--A total of 3291 women aged 50-64 years who were due to be called for breast screening for the first time. The analysis of predictors was based on a subsample of 1301, reflecting a response rate of 75% to interview and 36% to postal questionnaire. MAIN RESULTS--Attendance was 42% overall, and 70% in those who gave an interview or returned a questionnaire. There was little evidence for an interview effect on attendance. The main findings from the analysis of predictors are listed below. (These were necessarily based on those women who responded to interview/questionnaire and so may not be generalisable to the full sample.) (1) Sociodemographic factors: Women in rented accommodation were less likely to go for screening but other indicators of social class and education were not predictive of attendance. Age and other risk factors for breast cancer were unrelated to attendance, as was the distance between home and the screening centre. Married or single women were more likely to attend than divorced, separated, or widowed women, and black women had a higher than average attendance rate; however, neither of these relationships was found in the interview sample. (2) Health behaviours: Attenders were less likely to have had a recent breast screen, more likely to have had a cervical smear, more likely to go to the dentist for check ups, and differed from non-attenders with regard to drinking frequency. Exercise, smoking, diet change, and breast self-examination were unrelated to attendance. (3) Attitudes, beliefs, and intentions: The two best predictors were measures of the perceived importance of regular screening for cervical and breast cancer and intentions to go for breast screening. Also predictive were beliefs about the following: the personal consequences of going for breast screening, the effectiveness of breast screening, the chances of getting breast cancer, and the attitudes of significant others (the woman's husband/partner and children). Women who reported a moderate amount of worry about breast cancer were more likely to attend than those at the two extremes. CONCLUSIONS--Attenders and non-attenders differ in two broad areas: the health related behaviours they engage in and the attitudes, beliefs, and intentions they have towards breast cancer and breast screening. The latter are potentially amenable to change, and though different factors may operate among women who do not respond to questionnaires, the findings offer hope that attendance rates can be improved by targeting the relevant attitudes and beliefs. This could be done by changing the invitation letter and its accompanying literature, through national and local publicity campaigns, and by advice given by GPs, practice nurses, and other health professionals. It is essential that such interventions are properly evaluated, preferably in randomised controlled studies.