The key to auto-focusing of microscopic images is to design a high sensitivity focusing function. Due to uncertainty of the details among different microscopic images, traditional gradient functions are less sensitive to less detailed images. To solve the problem, an auto-focusing function, VarGrad, combining global and local gray-scale variation, is proposed. According to the characteristics of microscopic images, VarGrad function combines the gray variance function based on global gray-scale variation with the gray gradient function based on local gray-scale variation by using focusing windows. The VarGrad function exhibits high sensitivity regardless of image detail. We carry out quantitative evaluation of the proposed VarGrad function using two different peripheral blood cell image sequences with different image details. The experimental results show that three sensitivity indices, sharpness rate, steepness, and rate of change of sharpness, are improved by over 30% for detailed images and less detailed images, compared with that of several traditional typical focusing functions.