An efficient method is proposed for identification of linear noisy input-ou
tput systems. Central to this method is that the variances of the input and
output noises, which determine the bias in the ordinary least-squares (LS)
estimator, are estimated in the way of increasing the degrees of both the
denominator and the numerator of the system transfer function by one, but w
ith no need to evaluate the average LS errors. While achieving estimation u
nbiasedness, the proposed method exhibits algorithmic advantages over the L
S-based algorithms recently developed. Performance comparisons with other e
xisting estimation algorithms based upon computer simulations are given.