Global climate change influences the soil temperature and active layer freeze-thaw processes in Arctic permafrost soils. Increasing permafrost temperatures have been extensively studied, but changes in seasonal freeze-thaw processes have often been less regarded. This work aimed to examine the variability of the active layer freeze-back onset and duration at the Arctic research site on Samoylov Island at the Lena River Delta in northeastern Siberia. Samoylov is underlain by continuous permafrost and characterized by a polygonal tundra landscape. I analyzed soil temperature data from 2014 to 2020 with respect to the polygon’s microtopography. I estimated the timing of freeze-back at different depths in both the elevated rim and topographically lower center of the polygon and investigated the effect of snow cover accumulation on this process. During the study period, freeze-back started between September and October and ended between October and January. Between different depths, freeze-back onset was relatively stable while the duration showed more variability. Shallow depths refroze sooner than greater depths. I observed a shift between 7 d to 11 d towards later freeze-back onset at the polygon rim and center. While freeze-back duration was relatively stable in the upper part of the polygon rim, the polygon center and deeper layers at the rim showed great inter-annual variability. A notably long freeze-back duration occurred in 2015, 2016, and 2020. Meanwhile, freeze-back at the polygon center started later and took longer compared to the rim. Interestingly, the date of the first snowfall did not have a distinct impact on the freeze-back process. While the freeze-back onset was more closely connected to the air temperature, the duration was more likely influenced by snow depth, snow water equivalent (SWE), and volumetric water content (VWC). This work emphasizes the importance of gaining a more comprehensive understanding of freeze processes, their drivers, and the role of soil microtopography. It highlights the need to consider the variations in both process and drivers across different topographical features when assessing active layer conditions.