How can we develop a robot that can naturally interact with people? One approach is to build an android whose appearance and movements are very humanlike. People anthropomorphize objects for communication. Therefore, we can use androids as a testbed for studying human-human and human-machine interactions. This novel research approach is called android science. In android science, the methods of engineering and cognitive science converge. Roboticists attempt to build very humanlike robots based on knowledge from cognitive science, while cognitive scientists employ robots to verify cognitive hypotheses about human beings. This paper proposes the framework of android science, introduces the androids we have developed, and discusses conscious and subconscious human recognition of these androids as a fundamental issue for android science.