OBJECTIVE: To assess manifest bias in ageing effects, i.e. longitudinal changes due to unintended time effects or to selection. DESIGN: Mixed-longitudinal study in birth cohorts 1913-1918, with baseline measurements taken in 1988/1989 and repeated in 1993, including a short questionnaire in non-responders. SETTING: Full baseline and follow-up data were collected in nine towns in eight European countries including Belgium, Denmark, France, Italy, The Netherlands, Portugal, Spain and Switzerland. Incomplete data were available from towns in Portugal, Poland, Northern Ireland and Connecticut, USA. SUBJECTS: Using standardized methodologies data were collected from a random age-stratified sample of elderly men and women, including a total of 1221 re-invited subjects from nine towns and 210 newly-invited subjects from three towns in 1993. RESULTS: An overall retrieval of 50-74% of the former participants could be reached in towns that had previously participated (apart from one exception of 41%), where estimates of mortality varied from 10% to 18%. There was a tendency for healthy and active persons to have a higher participation rate than others, as was the case for high educated newly-invited subjects compared to lower educational classes. For most of the variables used in the analysis of period effects, no evidence of any undesirable period effect was found. In those instances that period effects showed up to be statistically significant, coinciding implausible cohort effects gave the impression that these were due to instability of the estimation procedure. CONCLUSIONS: Non-participants may be less healthy and active than the participants. Only very limited unconvincing evidence to suggest unintended time effects was observed. This confirms the high standards of the methodology and of measurements.