Consideration was given to estimation of the parameters of linear regression under arbitrary noise, that is, noise whose mean value is either unknown and other than zero, or a realization of a correlated random process, or defined by an unknown bounded determinate function.