Glomalin-related soil protein (GRSP), an operationally defined fraction of soil organic matter containing protein and various other components, is usually quantified using the colorimetric non-specific Bradford method. This method is limited by a short working range, a non-linear response and interference from co‐extracted compounds. These limitations hinder the exact quantification of the protein component. The aim of this study was to investigate the source of interference in the Bradford quantification of GRSP and propose several methodological improvements based on identified interferences. The easily extractable and total GRSP in five topsoils with contrasting texture, organic carbon content and land use were compared. Results showed that: (i) the extent of interference varied between different soils, (ii) the standard addition method overestimated the extent of inhibition, (iii) absorbance should be corrected for colour, (iv) use of the ratio of absorbances at 595 and 465 nm, A595/A465, is not recommended because it is sensitive to pH and dilution‐dependent absorbance at 465 nm, (v) although a quadratic fit to the protein calibration curve was better than the linear fit, it was not possible for the dilution method and (vi) estimation of protein content from the dilution curve of the soil extract appeared to be suitable as it integrates the often observed, and hitherto unexplained, effect of dilution on the calculated protein content of soil extracts and avoids artefacts because of the choice of protein spike and dilution.