We propose an object detection method using particle filters. Our approach estimates the probability of object presence in the current image given the history of observations up to current time. To do so, object presence is modelled by a two-state Markov chain, and the problem is translated into sequential Bayesian estimation which can be solved by particle filters. The observation density, required by the particle filter is based on selected discriminative Haar-like features that were introduced by Viola and Jones [7] for object detection in static images. We illustrate the approach on the problem of face detection. Experiments on real video sequences show the feasability of the approach.