We consider the sampled-data filtering problem by proposing a new performance criterion in terms of the estimation error covariance. An innovation approach to sampled-data filtering is presented. First, the definition of the estimation covariance e for a sampled-data system is given, then the sampled-data filtering problem is reduced to the Kalman filter design problem for a fictitious discrete-time system, and finally, an effective method is developed to design discrete-time Kalman filters in such a way that the resulting sampled-data estimation covariance achieves a prescribed value. We derive both the existence conditions and the explicit expression of the desired filters and provide an illustrative numerical example to demonstrate the directness and flexibility of the present design method.