In this paper, the authors consider the problem of tracking of a mobile target node in a Mobile Ad hoc NETwork (MANET) set-up. They find a gradient model alone is usually not very efficient, whereas a precise Markov model which includes transition probabilities is too hard to achieve. They propose a generic tracking framework for online tracking applications, by integrating a gradient model of the target's proximity and an online statistically estimated Markov model of the target's likely direction. They show PMBT achieves a short catching path with a high success rate. PMBT is a probabilistic online tracking algorithm that computes information utilities at each step, and then chooses the next step toward the target based on the maximum expected utility.