In this paper, we study stealthy false-data attacksagainst state estimators in power networks. The focus ison applications in SCADA (Supervisory Control and DataAcquisition) systems where measurement data is corrupted bya malicious attacker. We introduce two security indices for thestate estimators. The indices quantify the least effort neededto achieve attack goals while avoiding bad-data alarms in thepower network control center (stealthy attacks). The indicesdepend on the physical topology of the power network and theavailable measurements, and can help the system operator toidentify sparse data manipulation patterns. This informationcan be used to strengthen the security by allocating encryptiondevices, for example. The analysis is also complemented witha convex optimization framework that can be used to evaluatemore complex attacks taking model deviations and multipleattack goals into account. The security indices are finallycomputed in an example. It is seen that a large measurementredundancy forces the attacker to use large magnitudes in thedata manipulation pattern, but that the pattern still can be relatively sparse.