Abstract Background A recent review concludes that the agreement of data across ActiGraph accelerometer models for children and youth still is uncertain. The aim of this study was to evaluate the agreement of three generations of ActiGraph accelerometers in children in a free-living condition. Methods Sixteen 9-year-olds wore the ActiGraph AM7164, GT1M and GT3X+ simultaneously for three consecutive days. We compared mean counts per minute (mcpm) and time spent at different intensities from the three generations of monitors, and the agreement of outputs were evaluated by intra-class correlation coefficients (ICC) and Bland-Altman plots. Results The ICC for mcpm was 0.985 (95% CI = 0.898, 0.996). We found a relative difference of 11.6% and 9.8% between the AM7164 and the GT1M and AM7164 and the GT3X+, respectively. The relative difference between mcpm assessed by the GT1M and GT3X+ was 1.7%. The inter-generation differences varied in magnitude and direction across intensity levels, with the largest difference found in the highest intensities. Conclusion We found that the ActiGraph model AM7164 yields higher outputs of mean physical activity intensity (mcpm) than the models GT1M and GT3X+ in children in free-living conditions. The generations GT1M and GT3X+ provided comparable outputs. The differences between the old and the newer monitors were more complex when investigating time spent at different intensities. Comparisons of data assessed by the AM7164 with data assessed by newer generations ActiGraphs should be done with caution.