Ice-on date occurred significantly later over 1975-2009 at Dickie Lake, Ontario, while ice-off date showed no significant trend, differing from many other records in North America. We examined the ice phenology using 3 modelling approaches: a lake-specific regression model to derive a suite of local predictors; a regionally derived regression model to test larger-scale predictors; and a physically based, one-dimensional thermodynamic model. All 3 models were also applied to generate future ice cover scenarios. The local regression revealed air temperature to be an important predictor of ice phenology in our area, as reported elsewhere; however, reductions in wind speed and increases in lake heat storage over the last 35 years also contributed significantly to a delayed ice-on date. Ice-off dates were strongly correlated with the effects of warmer air temperatures but also influenced by increased snowfall and reduced wind speed. Thus, although changes in ice phenology were related to continental-scale changes in air temperature, they were also influenced by more localized climatic variables, and a careful examination of local events was needed for a complete assessment of ice phenology. Predictabilities of the regional regression model, which primarily relied on air temperature to predict phenology, and the physically based model were lower than the lake-specific local regressions, reinforcing the need for inclusion of local variables when greater accuracy is important. Finally, the 3 methods generated similar estimates of reductions in ice cover over the next 90 years, predicting a 40-50 day decrease in ice season length by 2100. © International Society of Limnology 2013.