Since the 1980s, AI and Law has attempted to capture legal expertise in computer programs. But what is this expertise? This paper reviews a number of approaches, from the 1980s to the present day, which represent different answers to this question. It argues that our notion, and understanding, of expertise has developed and improved over the decades. As yet, however, only a few rather specific aspects have been addressed in detail, in particular the move from intermediate predicates to legal consequences, and the distinguishing of precedents. Much more, including the moves from evidence to facts and from facts to intermediate predicates, awaits exploration.