PURPOSE: The present study aims to estimate the minimum number of accelerometer measurement days needed to estimate habitual physical activity (PA) among 6- (2010), 18- (2011) and 30- (2012) year-old participants, belonging to three population-based Brazilian birth cohorts. METHOD: PA was assessed by triaxial wrist-worn GENEActiv accelerometers and the present analysis is restricted to participants with at least 6 consecutive days of measurement. Accelerometer raw data were analyzed with R-package GGIR. Description of PA measures (overall PA, moderate-to-vigorous PA (MVPA), light PA (LPA)) on weekdays and weekend days were conducted, and statistical differences were tested with chi-squared and Kruskal-Wallis tests. Spearman Brown Formulae was applied to test reliability of different number of days of accelerometer use. RESULTS: Differences between week and weekend days regarding LPA, MVPA and overall PA, were only observed among 30-year-olds. Higher levels of MVPA (p = 0.006) and overall PA (p<0.001) were identified on weekdays. For overall PA, to achieve a reliability coefficient >0.70, two and three days of measurement were needed in adults and children, respectively. For LPA, a reliability coefficient >0.70 was achieved with five days in 6-year-old children, three days in 18-year-old young adults, and four days in 30-year-old adults. Considering MVPA, four days would be necessary to represent a week of measurement among all cohort groups. CONCLUSION: Our results show that four and five measurement days are needed to estimate all habitual PA constructs, for children and adults, respectively. Also, among 30-year-old adults, it is important to make efforts towards weekend days measurement.