A new particle tracking method for two-dimensional particle tracking velocimetry (2-D PTV) is proposed. The method, named as a 'triple pattern matching algorithm', integrates two different kinds of information that can be derived from three consecutive frames of particle images. One is geometrical similarity recognized between local particle distribution patterns. The similarity is quantified by the overlapping area of particle images. A simple technique is devised to deal with possible deformation of particle patterns owing to the local fluid motions. The other kind of information used is predictability of particle motions. Analysis of Lagrangian particle trajectories shows that the use of three (or more) consecutive frames has definite advantage over the use of just two frames. It is also shown that the optimum tracking parameter is determined by the ratio of the mean and fluctuating velocity components. The present method is verified by using synthetic data with which laminar Couette flows and solid rotation flows are simulated. Moreover, measurement of a stirred water flow is carried out to examine the present method experimentally.