A class of parametric tests for heteroscedasticity in linear models is discussed. For models with nonstochastic regressors, new exact tests within this class are suggested which utilize existing tables of the distribution of the von Neumann ratio and of the Durbin-Watson bounding ratios. "Bound tests" for heteroscedasticity in least squares regression are proposed. A rigorous treatment of tests within this class for heteroscedasticity in the errors of structural relations in dynamic simultaneous equations models is provided.