Abstract: In a method for the phase correction of nuclear magnetic resonance signals given a pulse sequence wherein image echos are acquired under readout gradients of alternating operational sign, first and third reference echos (S.sub.1.sup.+, S.sub.3.sup.+) are acquired under sub-pulses of the readout gradient with a first polarity and a second reference echo (S.sub.2.sup.-) is acquired under a sub-pulse of the opposite polarity. An interpolation echo (S.sub.2.sup.+) that fictitiously occurs at the echo time of the second reference echo (S.sub.2.sup.-) is formed by interpolation from the first and third reference echo (S.sub.1.sup.+, S.sub.3.sup.+). Correction data are determined by comparing the second reference echo (S.sub.2.sup.-) to the interpolation echo (S.sub.2.sup.+).