The patterns of Human Epithelial type 2 (HEp-2) cell provide useful information for the diagnosis of systemic autoimmune diseases. However, the recognition of cell patterns requires manual annotation by experienced physicians, which is subject to inter-observer variability. Therefore, an automatic diagnosis system is desirable. As the crucial pre-processing step for cell pattern recognition, the performance of cell segmentation is crucial. In this paper, a novel adaptive local thresholding approach is proposed to solve the issue. The approach divides cell images into overlapping sub-images and applies adaptive threshold estimator to each of them. The ICPR 2014 HEp-2 cell datasets are employed to assess the segmentation performance of our framework. The results show that the system achieves an average segmentation accuracy of 66.95%, which outperforms the typical thresholding approaches.