Abstract: In this report we describe the models with which we experimented to predict rodent behavior from video recordings. Automatic recognition of rodent behavior from video is a desirable tool in behavioral studies since it can significantly reduce the required human effort and simultaneously reduce the dependence of results on particular human observers. In our research we considered several variants of Hidden Markov Models and compared results against a simple logistic discriminant classifier that ignores the correlation of behaviors between successive frames. For a selection of four behaviors a correct classification rate around 75% is obtained.