Advanced electron beam dose calculation models for radiation oncology require as input an initial phase space (IPS) that describes a clinical electron beam. The IPS is a distribution in position, energy and direction of electrons and photons in a plane in front of the patient. A method is presented to derive the IPS of a clinical electron beam from a limited set of measured beam data. The electron beam is modelled by a sum of four beam components: a main diverging beam, applicator edge scatter, applicator transmission and a second diverging beam. The two diverging beam components are described by weighted sums of monoenergetic diverging electron and photon beams. The weight factors of these monoenergetic beams are determined by the method of simulated annealing such that a best fit is obtained with depth-dose curves measured for several field sizes at two source-surface distances. The resulting IPSs are applied by the phase-space evolution electron beam dose calculation model to calculate absolute 3D dose distributions. The accuracy of the calculated results is in general within 1.5% or 1.5 mm; worst cases show differences of up to 3% or 3 mm. The method presented here to describe clinical electron beams yields accurate results, requires only a limited set of measurements and might be considered as an alternative to the use of Monte Carlo methods to generate full initial phase spaces.