The bias compensated least squares approach for errors-in-variables model identification is examined in a new framework, where it is allowed to prefilter the observed input-output data prior to the estimation process. A statistical analysis of the estimation algorithm is presented. Subsequently, it is shown how these prefilters and the weighting matrix can be tuned in order to optimize the estimation accuracy. According to the numerical simulation results, the covariance matrix of the estimated parameter vector is very close to the Cramer-Rao lower bound for the estimation problem.