The accuracy of determining the point of force application with piezoelectric force plates, as specified by the manufacturer, is lower than needed for certain applications. The purpose of this study was to evaluate the accuracy of a commonly used plate (KISTLER type 9287) and to improve it by proposing a correction algorithm. Forces were applied to a wooden board, supported in one corner by a stylus that rested on the force plate. To determine the influence of position and magnitude of the force vector, the stylus was placed on 117 different locations, and calibrated masses were used to exert vertical forces between 0 and 2000 N. To determine the influence of loading rate, dynamic tests were performed in which a subject ran across the board. In static tests at a given stylus position with actual coordinates x (short axis) and y (long axis), it was found that the calculated coordinates x ̂ and y ̂ of the point of force application had virtually constant values at forces above 1000 N. In dynamic tests, oscillations could occur in x ̂ and y ̂ with an amplitude of more than 20 mm. When these were avoided or removed by filtering, static and dynamic tests at a given stylus position showed the same values for x ̂ and y ̂ at forces above 1000 N. Across stylus positions, the errors x ̂-x and y ̂-y (measured at 1600 N) ranged from -20 to +20 mm. The average over 117 points of the absolute errors | x ̂-x| and | y ̂-y| amounted to 3.5 and 6.3 mm, respectively (mean values of three plates). Based upon the observed error distribution, a correction algorithm was selected. After correction, the errors ranged between -10 and +10 mm, and | x ̂-x| and | y ̂-y| were reduced to 1.3 and 1.6 mm, respectively. The algorithm may be applied by other users of the force plate tested here to improve their accuracy in determining the point of force application.