Gridding, which has a large impact on the identification of differentially expressed genes, is the first and key step for microarray image analysis. Most gridding methods are semi-automatic or require parameter preset. In this paper, an improved method was proposed for rapid and accurate gridding compared to the mathematical morphology based method. First, the image quality was enhanced by using the logarithm transformation. Then, an optimal threshold was gained based on Otsu method. Experiments on microarray images drawn from SMD and GEO prove that our method is fully automatic and need no parameter, with high accuracy in the presence of lots of noise.