Frequency dependent structure is not uncommon in observations of extended radio sources. Where a source consists of two components with different spectra at an angular separation considerably less than the resolution of the telescope, the centroid position will vary systematically with frequency. A well-known example is 3C 273 though the effect in this case is quite small. With beams of a few minutes of arc, larger position differences could occur between measurements made at widely separated frequencies without the source being noticeably resolved. In a preliminary study of a sample of 120 4C radio sources, Munro and Hoskins found 17 cases of discrepancies in the range 45 to 100 seconds of arc between positions measured at 408 MHz with the Molonglo cross (beam width ~3′ arc) and at 2700 MHz with the Parkes 64 m telescope (beam width ~7′ arc) by Wills and Bolton. These separations were considered to be significant in terms of the quoted errors which were typically 10 to 15″ arc in each coordinate for both sets of measurements. It was pointed out that to explain the results on the basis of a double source model it would be necessary for the two components to differ in spectral index by ~1.