This paper proposes a new method for identifying structural parameters in models with endogenous or mismeasured regressors. The method is useful when traditional identification sources like instrumental variables or repeated measurements are not available. Identification is achieved through heteroskedasticity, which is a feature of many models with endogeneity or measurement error. The method also applies to semiparametric partly linear models, and set identification bounds are derived when point identification assumptions fail. An empirical application estimates Engel curves, showing that the method yields results similar to those obtained using traditional instruments. The paper considers triangular and simultaneous systems, where identification comes from the uncorrelation of regressors with the product of heteroskedastic errors. For triangular systems, identification is obtained by assuming that the covariance of errors with regressors is non-zero. For simultaneous systems, identification is achieved by assuming that the covariance of errors with a subset of regressors is zero. These assumptions allow for the estimation of structural parameters using two-stage least squares or generalized method of moments. The paper also discusses empirical applications, including a study of Engel curves where total expenditures may be mismeasured. The results show that the method provides estimates similar to those obtained using traditional instruments. The paper also reviews related literature, showing that the method is applicable in various settings where traditional instruments are not available. The method is shown to be effective in models with classical measurement error, unobserved common factors, and other applications. The paper concludes with estimation methods and extensions, including set identification and nonlinear models.This paper proposes a new method for identifying structural parameters in models with endogenous or mismeasured regressors. The method is useful when traditional identification sources like instrumental variables or repeated measurements are not available. Identification is achieved through heteroskedasticity, which is a feature of many models with endogeneity or measurement error. The method also applies to semiparametric partly linear models, and set identification bounds are derived when point identification assumptions fail. An empirical application estimates Engel curves, showing that the method yields results similar to those obtained using traditional instruments. The paper considers triangular and simultaneous systems, where identification comes from the uncorrelation of regressors with the product of heteroskedastic errors. For triangular systems, identification is obtained by assuming that the covariance of errors with regressors is non-zero. For simultaneous systems, identification is achieved by assuming that the covariance of errors with a subset of regressors is zero. These assumptions allow for the estimation of structural parameters using two-stage least squares or generalized method of moments. The paper also discusses empirical applications, including a study of Engel curves where total expenditures may be mismeasured. The results show that the method provides estimates similar to those obtained using traditional instruments. The paper also reviews related literature, showing that the method is applicable in various settings where traditional instruments are not available. The method is shown to be effective in models with classical measurement error, unobserved common factors, and other applications. The paper concludes with estimation methods and extensions, including set identification and nonlinear models.