A method for suppressing fat in fast spin-echo imaging with the three-point Dixon technique is described. The method differs from the three-point Dixon method used in conventional spin-echo imaging in that the readout gradient instead of a radio-frequency pulse is shifted. This method preserves the Carr-Purcell-Meiboom-Gill nature of the fast spin-echo sequence and hence is less sensitive to magnetic field inhomogeneities and resonance frequency mistiming. As in the original three-point Dixon technique used in conventional spin-echo imaging, three acquisitions are required to estimate the field inhomogeneity and completely separate fat and water. The extra time required is not excessive considering that the fast spin-echo method is frequently applied with multiple signal acquisition. Also, this technique achieves an expected signal-to-noise ratio comparable to 2.67 signal acquisitions, which is approximately 94% of the signal-to-noise ratio obtained with three signal acquisitions. The method is demonstrated with applications to phantoms and a human volunteer.