With the coming era of high-technology, the intelligent factory has become mainstream for each firm to follow. How to reduce the labor cost is grown into a crucial issue for industrial field. Since integrated circuit (IC) package defect inspection is a complicated and time consumptive task which always charge amount of human resource for companies, in this paper, we propose an automatic IC crack detection method to save the labor cost. First, we develop a high-precision region of interest (ROI) detecting method to reduce the overkill rate. Then, we conduct pre-processing step, filtering the noise in the image to enhance the feature of crack. Finally, we established an IC crack recognition method via variance analysis algorithm to classify the pass IC and failed IC. Experimental results show that we can achieve 83% of accuracy rate with our method.