The design and operation of membrane nanofiltration processes requires the use of good numerical predictive methods and prior characterisation of membrane properties. The best methods currently available are based on the use of the extended Nernst-Planck equation and a characterisation of the membrane in terms of the effective pore radius (r(p)), the effective charge density (X-d) and the ratio of effective membrane thickness to membrane porosity (Delta x/A(k)). When the ranges of such parameters for commercially available nanofiltration membranes are known, it becomes possible to use the numerical predictive method to choose the membrane best suited to a particular process requirement. The present paper considers such an analysis for the processing of dye/salt/water streams. It is shown that if the requirement is the removal of salt through diafiltration as part of the production process, then it is best to choose a membrane with a moderate value of r(p) (matching the hydrodynamic radius of the dye), a low value of X-d and a low value of Delta x/A(k). However, if the requirement is the removal of colour from a waste rinse stream, then the best results will be obtained by choosing a membrane with a high value of r(p) and a high value of X-d.