Can CEM software ever be validated?

Validation of computational electromagnetics software has been a key topic for code developers for many years. The question arises: can software ever be validated meaningfully? Two cases are examined. Case 1: an electromagnetic code has been written to solve Maxwell´s equations using the finite element method, including effects of eddy current losses and displacement currents (i.e. the full set of equations). The software is compared against a number of test problems, each having an analytic solution. It is shown that any required level of accuracy can be achieved by refining the mesh. Is this software therefore fully tested? Case 2: an electromagnetic code has been written to solve a subset of Maxwell´s equations, neglecting displacement currents. It is used to model an induction heater and comparison is made with measured results. The results are found to be close to measurement but there are some differences. Is this software therefore in error? This paper discusses these two approaches to software validation in more detail and makes some recommendations.