Accurately estimating winter surface mass balance on glaciers is central to assessing glacier health and predicting glacier run-off. However, measuring and modelling snow distribution is inherently difficult in mountainous terrain. Here, we explore rigorous statistical methods of estimating winter balance and its uncertainty from multiscale measurements of snow depth and density. In May 2016, we collected over 9000 manual measurements of snow depth across three glaciers in the St. Elias Mountains, Yukon, Canada. Linear regression, combined with cross-validation and Bayesian model averaging, as well as ordinary kriging are used to interpolate point-scale values to glacier-wide estimates of winter balance. Elevation and a wind-redistribution parameter exhibit the highest correlations with winter balance, but the relationship varies considerably between glaciers. A Monte Carlo analysis reveals that the interpolation itself introduces more uncertainty than the assignment of snow density or the representation of grid-scale variability. For our study glaciers, the winter balance uncertainty from all assessed sources ranges from 0.03 to 0.15 m w.e. (5–39%). Despite the challenges associated with estimating winter balance, our results are consistent with a regional-scale winter-balance gradient.