Validation of sea-ice models is difficult because observations of sea ice in high latitudes are scarce and the metrics commonly used to measure model skill are not always adequate. For instance, a model configuration can produce good agreement between some simulated sea ice quantity and observations, but perform poorly when comparing another quantity. We develop a variance-based distance metric suitable to assess the skill of sea ice models when multiple output variables (e.g. concentration, thickness, age), or uncertainties in both model predictions and observations are to be considered. The validation metric we implemented is a robust statistic that measures the normalized distance between simulated and observed quantities of interest such as sea ice thickness, concentration and draft. The distance metric normalizes the squared differences between model results and observations by a variance measurement associated with the model, the observations, or both. These features allow the integration of different quantities of interest, and observational and model uncertainties in the validation assessment. The proposed methodology uses the fact that the distance metric is a gamma-distributed statistic, which can be used to assert whether or not simulated quantities match observations, and to use probability as the yardstick to measure model skill. This methodology is therefore a statistically robust framework for validation of sea ice models.