In this paper we present an off-line Kalman filter approach to remove transcranial magnetic stimulation (TMS)-induced artifacts from electroencephalographic (EEG) recordings. Two dynamic models describing EEG and TMS signals generation are identified from data and the Kalman filter is applied to the linear system arising from their combination. The keystone of the approach is the use of time-varying covariance matrices suitably tuned on the physical parameters of the problem that allow to model the nonstationary components of the EEG-TMS signal. This guarantees an efficient deletion of TMS-induced artifacts while preserving the integrity of EEG signals around TMS impulses. Experimental results show that the Kalman filter is more effective than stationary filters (Wiener filter) for the problem under investigation.