A new calibration method has been developed to improve the accuracy of chromaticity coordinates obtained from a tristimulus colorimeter for color displays. Matrix methods such as the one recommended by ASTM are well known for this purpose, but they may fail to work as expected due to experimental noise and errors. As these matrix methods are based on tristimulus values, the accuracy of the luminance measurement affects the accuracy of the corrected chromaticity. This new method utilizes x, y values only, and is independent of y values. Thus, in principle, it eliminates errors due to luminance measurement variations. A correction matrix is obtained from the x, y values of three primary colors and a white color of a display, measured by the target instrument and a reference instrument. A computer simulation was conducted to evaluate the effect of random noise in y. Experiments were conducted using a commercial tristimulus colorimeter and a spectroradiometer, measuring 14 colors of a CRT display. The results show noticeable improvement in chromaticity accuracy over the current practice.