I like Craig's original approach because performance is good and it is straightforward. Another important consideration is many, if not most organizations that I am aware of tend to follow the 'one-employee-one-manager' model. I utilize essentially the same algorithm with three differences: 1) I encapsulate all of the code within the stored procedure, which simply outputs an ordered four-column dataset (SortName [Last, First], Employee ID, SupervisorID, Level), 2) I include the Level column (integer), which indicates how many levels down from the specified manager each employee is in the hierarchy, and 3) I have an additional input parameter for Level, which allows me to easily return only the direct reports for the manager (or some other number of levels deep that I may be interested in at the moment). Of course, my method makes certain assumptions about how the names will be used since I am controlling the name format (we do a lot of dropdown lists and reports using the output.)