In digital color printing, printer gamuts are often modeled as functions from CMY space into a device independent color space such as CIE XYZ tristimulus values. To render large raster images across devices, these gamut functions must be evaluated and inverted very efficiently; such performance can be provided only if the gamut function is represented as a look-up table, and evaluated by interpolation. The most common interpolation method uses data on a rectilinear grid, sometimes based on division of the cells into tetrahedra. It is not always possible to use a rectilinear scheme: available gamut measurements may not lie on such a grid, and the inverse of a gamut function sampled on a rectilinear grid does not take this form. Based on ideas developed for the numerical solution of partial differential equations, this paper develops a general tetrahedral interpolation technique that works efficiently with nonuniform data. The technique is shown to extend easily into higher-dimensional spaces.