In order to build an elastic-plastic model of soil behaviour it is necessary to define a yield function and a plastic potential. It is often assumed that these coincide and the so-called normality rule is obeyed. There is, however, much evidence that for soils the assumption of normality will not permit satisfactory predictions of soil behaviour to be made.Whereas the determination of the plastic potential from experimental studies is not difficult it is less easy to determine the form of the yield function. The indefinite yield points that are observed render the use of stress probe experiments somewhat inaccurate for this purpose.Here an alternative approach is proposed based on theoretical considerations. The form of the yield function is deduced for one set of experimental data for sand and also for one set of data for clay.