Printed documents often contain text and halftones on the same page. When processing these documents electronically, it is important to separate halftones from text, since their processing requirements differ. We present an algorithm that extracts both grayscale and color halftones from other information in the page. Halftones are commonly produced as a pattern of dots on a fixed lattice. The halftone dot pattern forms an "invisible texture", producing a spectral component distinct from the other information in the page. We show that a single circularly-symmetric bandpass filter, tuned to this spectral component, can be used to extract halftones from text, even in the presence of image rotations.