Errors-in-Variables (EIV) models consider the presence of additive errors on the measures of all measurable attributes of a process. Traditional identification procedures for these processes rely on the properties of the covariance matrix of the observations and on its decomposition into the sum of a matrix associated with the (unknown) noiseless sequences and of the observation noise covariance matrix. This paper makes reference, in a behavioural framework, to the observed noisy trajectories and performs a decomposition of these trajectories into a regular part, defininig the associated behaviour of the process, and a noise part. The Monte Carlo simulations that have been performed show that the proposed approach leads to accurate estimates of both the system parameters and the noise variances.