In this paper the frequency domain version of the Frisch identification scheme is applied to identify parameters of the continuous-time model of an induction motor. A formulation of the identification problem in the errors-in-variables framework is given, in particular this formulation allows handling of periodic signals affected by noises with stochastic properties. A new approach, based on Bilinear Matrix Inequalities, is introduced to estimate noise variances of measured signals in the Frisch scheme. Simulations and experimental results are reported to show the properties of the proposed approach.