A vision coordinate measurement method based on microscopic image mosaic technique is proposed, that is to say, the coordinate measurement is implemented with spacial coordinate information in image itself. According to fine texture feature on small parts images captured by the measuring microscope, a cross-correlation matching algorithm based on the point map was presented to mosaic sequence images provided with partial overlap. By means of a rigorous system calibration using glass standard rule, mosaicing errors were calculated and analyzed on different microscopic magnification. Then the measuring example of small part dimensions was illuminated, which shows that the algorithm is effective. Several experiment studies show that relative error is superior to 1% and resolution is less than 1 micrometer.