Two efficient and exact algorithms for computing the N best sentence hypotheses in continuous speech recognition are studied and compared. Both procedures are first stated as algorithmic solutions to the N shortest paths problem in graphs. The first algorithm is based on the general A* search strategy. The second one is based on a recursive procedure that enumerates shortest paths ending at a given node and is referred to as the Recursive Enumeration Algorithm.