Three different classes of bias-compensating least squares identification methods are compared, and shown to be identical. It is also discussed how the user parameters in the classes can be chosen to achieve optimal accuracy of the parameter estimates. (C) 2013 Elsevier Ltd. All rights reserved.