The use of satellite snow cover information to estimate parameters in grid distributed snowmelt models (GSMs), requires an error model quantifying the severity of a simulation-observation mismatch. This error model is the basis for the likelihood in statistical (Bayesian or frequentistic) estimation, or for an objective function in parameter calibration. As a part of research directed to construct an error model, this paper compares simulated snow coverage estimates from an energy-sum snowmelt model to MODIS observations. An important aspect of this comparison is to determine if the errors appear severely biased, or suggest that assumptions are not met. This is likely to cause over-conditioning on the vast amount of information supplied. The error analysis is split in two subtasks; errors of MODIS data compared to a “ground truth” from Landsat ETM+, and errors of an energy-sum GSM compared to the MODIS data. It is shown that a non-thresholded variable, such as the Normalised Difference Snow Index (NDSI), has desirable advantages as basis for the error model, compared to Snow Covered Area (SCA). NDSI is more homoscedastic, and provides more stable statistics when SCA is close to 0 or 1. The GSM simulation errors have a standard deviation of between 0.2 and 0.4, compared to 0.1 for the MODIS measurement. Analysis of the GSM simulated error structure versus MODIS NDSI suggests serious weaknesses in the gridded snowmelt model, and is discussed in this context rather than proposing an error model.