Context. To properly determine the role of magnetic fields during massive star formation, a statistically significant sample of field measurements probing different densities and regions around massive protostars needs to be established. However, relating Zeeman splitting measurements to magnetic field strengths needs a carefully determined splitting coefficient.Aims. Polarization observations of, in particular, the very abundant 6.7 GHz methanol maser, indicate that these masers appear to be good probes of the large scale magnetic field around massive protostars at number densities up to n(H2) approximate to 10(9) cm(-3). We thus investigate the Zeeman splitting of the 6.7 GHz methanol maser transition.Methods. We have observed of a sample of 46 bright northern hemisphere maser sources with the Effelsberg 100-m telescope and an additional 34 bright southern masers with the Parkes 64-m telescope in an attempt to measure their Zeeman splitting. We also revisit the previous calculation of the methanol Zeeman splitting coefficients and show that these were severely overestimated making the determination of magnetic field strengths highly uncertain.Results. In total 44 of the northern masers were detected and significant splitting between the right-and left-circular polarization spectra is determined in >75% of the sources with a flux density >20 Jy beam(-1). Assuming the splitting is due to a magnetic field according to the regular Zeeman effect, the average detected Zeeman splitting corrected for field geometry is similar to 0.6 m s(-1). Using an estimate of the 6.7 GHz A-type methanol maser Zeeman splitting coefficient based on old laboratory measurements of 25 GHz E-type methanol transitions this corresponds to a magnetic field of similar to 120 mG in the methanol maser region. This is significantly higher than expected using the typically assumed relation between magnetic field and density (B proportional to n(H2)(0.47)) and potentially indicates the extrapolation of the available laboratory measurements is invalid. The stability of the right-and left-circular calibration of the Parkes observations was insufficient to determine the Zeeman splitting of the Southern sample. Spectra are presented for all sources in both samples.Conclusions. There is no strong indication that the measured splitting between right-and left-circular polarization is due to non-Zeeman effects, although this cannot be ruled out until the Zeeman coefficient is properly determined. However, although the 6.7 GHz methanol masers are still excellent magnetic field morphology probes through linear polarization observations, previous derivations of magnetic fields strength turn out to be highly uncertain. A solution to this problem will require new laboratory measurements of the methanol Lande-factors.