Given two pictures of a scene taken by different sensors or at different times, one way of matching the two pictures is to extract a set of distinctive local features from each, and then match the resulting point patterns. This paper investigates the sensitivity of point pattern matching to various types of noise and distortion, including omissions and additions, random walks, rotation and rescaling, as well as the use of different feature detection operators to extract the points. The effects of additional information (e.g., feature types) in overcoming the effects of noise is also studied.