We present a new technique to estimate the abundance of water-ice in the Martian permafrost using Phoenix Lander Surface Stereo Imager (SSI) multispectral data. Past work estimated this abundance by employing radiative transfer methods to model the spectra of ice and regolith mixtures using the optical constants of water-ice and a Martian-analogue soil. Our technique removed the uncertainty of using an analogue (or of estimating a regolith composition) by deriving the optical constants directly from observations of icy regolith acquired before and after sublimation of the ice at similar viewing geometries. Laboratory spectral measurements of known mixtures of water-ice and dry soil at controlled viewing geometries allowed us to constrain the uncertainty in our technique. We found that model fits across the observational uncertainty will contain the actual water-ice fraction. We then applied the technique to Phoenix SSI observations of Snow White Trench and Dodo-Goldilocks Trench. For the Snow White Trench, we estimated water-ice abundances consistent with pore-fill in the near-surface ice, consistent with atmospheric diffusion. For the Dodo-Goldilocks Trench, we estimate water-ice abundances larger than pore-fill would allow. These results have implications to our understanding of the history of water-ice on Mars and the role of the regolith at high latitudes as a reservoir of atmospheric H2O. •We model spectra of regolith-ice mixtures to estimate ice content of samples.•We derive regolith optical constants from spectra of non-icy regolith.•We demonstrate this technique predicts ice fraction in laboratory experiments.•We apply the method to regolith in trenches excavated by the Mars Phoenix lander.