The procedure for polarisation measurements with radio synthesis telescopes is fraught with uncertaincies and shaky assumptions. Therefore, their accuracy will be limited to a few percent at best. It is investigated here whether it will be possible to improve this accuracy by a factor 10-100 with the WSRT, which has excellent polarisation characteristics, very small closure errors, and redundant spacings. The latter can be used to calibrate the instrument in a way which is less dependent upon assumptions about the polarisation of the calibrator source. They also cause more reliable convergence of the SELFCAL model, by reducing the number of free parameters. Although the investigation is not yet finished, the tentative conclusion is that it may be possible, with crossed dipoles, to derive completely model-independent values for Q/I, U/I and V/I.