For descriptor systems with measurement output noises (input disturbances may exist at the same time), a new descriptor estimator technique is developed. The necessary and sufficient condition for the existence of the present estimator is derived, and a systematic design approach is addressed. The effect of uncertainties is decoupled completely, and the asymptotic estimates of the descriptor system state and the output noise are obtained simultaneously. Furthermore, a normal state/disturbance estimator is also given. For a class of nonlinear descriptor systems with both output noises and input uncertainties, a nonlinear descriptor estimator is derived by using the proposed design approach, together with the linear matrix inequality technique. The existence and convergence of the nonlinear descriptor estimator is proven. The asymptotic estimates of the descriptor nonlinear system state and the output noise are obtained at the same time. The present estimators are applied to the sensor fault diagnosis, and hence the sensor fault can be estimated asymptotically. Finally, two numerical examples are included to illustrate the proposed design procedures and applications.