A preferable alternative to ordinary least squares (seemethod of least squares) when estimating the parameters of a model using independent observations whose values are known to vary in accuracy in a specified way. See alsogeneralized least squares.