[A regression model in which the disturbances exhibit a certain type of heteroscedasticity is considered. Maximum likelihood methods of estimation are developed and compared with the two-step estimation procedure. A likelihood ratio test for heteroscedasticity is suggested.]