Screw disassembly is a core operation in recycling electronic wastes (E-wastes), including mobile phone mainboards (MPMs). Currently, screw disassembly in most cases is still conducted manually which is inefficient and may adversely affect the health of workers. With the continuous development of intelligent manufacturing, a series of screw location methods have been designed to realise automated screw disassembly for various E-wastes. However, these methods cannot identify and classify tiny screws on complex MPMs. To overcome this limitation and expand the application domain of intelligent manufacturing, an accurate screw detection method, incorporating Faster R-CNN (high-performance deep learning algorithm) and an innovative rotation edge similarity (RES) algorithm, is proposed. In the experiments, the proposed method achieved a minuscule location deviation of 0.094 mm and satisfactory classification accuracy of 99.64%. The success rate and speed of automated screw disassembly for MPMs reached up to 90.8% and 4.98 s per screw, respectively. These results obtained from independently designed platforms confirm the practicality of the proposed method. However, incompleteness of detected screw groove edges can hamper the performance of RES; additionally, the computing speed of RES is currently unsatisfactory. In the future, solutions to the aforementioned drawbacks will be pertinently obtained.