A method for performing an evasive maneuver using a control device. A lane model is created for a time interval on the basis of received measurement data from at least one sensor. An incident corridor, within which an evasive maneuver is possible within the time interval, is ascertained based on the received measurement data and/or the created lane model. An evasion path is planned within the ascertained incident corridor. The control commands for longitudinal guidance and/or transverse guidance of a vehicle for traveling along the evasion path are generated on receipt of a trigger incident within the time interval. A stable lane model for at least one time interval is saved at least temporarily. A stable lane model from a past time interval is used to ascertain the incident corridor if a lane model from a current time interval is unstable.