Pose estimation involves reconstructing the configura- tion of a moving body from images sequences. In this paper we present a general framework for pose esti- mation of unknown objects based on Shafer's eviden- tial reasoning. During learning an evidential model of the object is built, integrating different image fea- tures to improve both estimation robustness and pre- cision. All the measurements coming from one or more views are expressed as belief functions, and com- bined through Dempster's rule. The best pose esti- mate at each time step is then extracted from the resulting belief function by probabilistic approxima- tion. The choice of a sufficiently dense training set is a critical problem. Experimental results concerning a human tracking system are shown.