Amplitude comparison is an efficient method to find the direction of arrival (DOA) of a radio frequency (RF) emitter. However, the DOA accuracy obtained by this method is usually poor due to beam pattern distortion. This paper provides a localization and beam pattern auto-calibration method, which estimates a RF emitter position and antenna beam pattern distortion simultaneously. The problem is formulated as a nonlinear dynamic estimation problem with an augmented state of emitter position and antenna distortion parameter. The state is directly estimated from the amplitude outputs of a two-element array. A multiple model unscented Kalman filter (MM-UKF) consisting of multiple initial states is proposed. It can adapt to different beam pattern distortion levels automatically. The algorithm is tested on a simulated antenna system. The results show that the MM-UKF outperforms the single model UKFs, and provides stable root mean square errors (RMSE) on different distortion levels.