A method of MIP mapping. Since the texels are almost always larger or smaller than the screen pixels, it finds two MIP-maps whose texels are closest in size to the screen pixels: one with larger texels, and the other with smaller texels. For each of the two MIP-maps, it then interpolates the four texels that are the nearest to each screen pixel. In the final step it averages between the two MIP results to render the final screen pixel.
Trilinear mip-mapping requires more than twice the computational cost of bilinear filtering, but the textures are filtered very nicely, with a clean result.