Please look again . The code I posted does an index scan on the City index. The primary key is CustomerID. The statistics must tell SQL Server that scanning the City index to find CustomerIDs is cheaper than doing a full table scan.