Autofocusing is the key of numerical reconstruction in digital holographic microscopy. Based on Fresnel transform algorithm, the performance of six typical autofocusing evaluation functions are studied experimentally. A method which only uses part of the reconstruted image as autofocusing area is proposed to decrease the calculating time. The results show that these functions have better unimodality, same adjusting focus range and accuracy, but the functions related with gradient are too slow to use in the numerical quasi-real time reconstruction. Fourier spectrum function is the optimum one for its minimum calculating time. The adjusting focus time can be decreased greatly by choosing a part of the reconstructed image as calculating area. For low signal-to-noise ratio hologram, the autofocusing accuracy and image quality can be improved by applying Wiener self-adaptive filter.