The thesis is related to modeling uncertainties in magnetic properties of ferromagnetic materials which occur due to, e.g., manufacturing imperfections. The modeling procedure is based on the random field approach. The Karhunen-Loève expansion is used to approximate random fields with a minimal number of random variables in mean square sense. Two main cases are covered. First, magnetic hysteresis is taken into account and uncertainties are modeled by introducing the concept of multivariate random field. Second, spatial uncertainties of the magnetic behavior law are also treated. Benchmark examples such as a single phase transformer and a combined function magnet are used to illustrate how to propagate uncertainties from the input material data to the relevant output data.