Under-ice dissolved oxygen (DO) metabolism and DO depletion are poorly understood, limiting our ability to predict how changing winter conditions will affect lake ecosystems. We analyzed under-ice DO dynamics based on high-frequency (HF) data at two depths (5 and 25 m) for three winters (January-March 2014, 2015, and 2016) in oligotrophic Lake Tovel (1178 m above sea level; maximum depth 39 m). Specifically, we assessed diel metabolic rates based on HF data of DO, temperature, and light for winter 2016 and seasonal DO depletion rates based on HF data of DO for all three winters. For 2016, calculations of metabolic rates were possible only for 34% and 3% of days at 5 and 25 m, respectively; these metabolic rates generally indicated net heterotrophy at both depths. Low success in modeling metabolic rates was attributed to low diel DO variability and anomalous diel DO patterns, probably linked to under-ice physical processes. Seasonal DO patterns for the three winters showed increasing, decreasing, or stable DO trends at 5 m while at 25 m patterns always showed decreasing DO trends but with different rates. Our multiyear study permitted us to hypothesize that the observed intraannual and interannual differences in DO depletion can be attributed to variable snow cover determining the penetration of radiation and thus photosynthesis. This study brings new insights to DO dynamics in ice-covered systems, highlights the challenges linked to under-ice lake metabolism, and advocates for a modeling approach that includes physical processes.