In a method and apparatus for recording a magnetic resonance data set of a target region of an object, wherein the target region contains at least one interfering object with a susceptibility difference from the rest of the target region that influences the homogeneity of the basic magnetic field, in particular a metal object and/or an air inclusion, in addition to a first raw data set of the target region recorded without additional dephasing, at least one further raw data set of the target region is recorded that corresponds to a raw-data specific additional dephasing of the spins in the target region. For each image point of the magnetic resonance data set, the maximum value raw data of the corresponding image points of all raw data sets in spatial domain are selected as magnetic resonance data.