AbstractEstimating a glacier’s volume by inferring properties at depth (e.g. bed topography or basal slip) from properties observed at the surface (e.g. area and slope) creates a calculation instability that grows exponentially with the size of the glacier. Random errors from this inversion instability can overwhelm all other sources of error and can corrupt thickness and volume calculations, unless problematic short spatial wavelengths are specifically excluded. Volume/area scaling inherently filters these short wavelengths and automatically eliminates the instability, while numerical inversions can also give stable solutions by filtering the correct wavelengths explicitly, as is frequently done when ‘regularizing’ a model. Each of the scaling and numerical techniques has applications to which it is better suited, and there are trade-offs in resolution and accuracy; but when calculating volume, neither the modeling nor the scaling approach offers a fundamental advantage over the other. Both are significantly limited by the inherently ‘ill-posed’ inversion, and even though both provide stable volume solutions, neither can give unique solutions.