Retrieving row count without using Count function

There are actually two problems with the query. The first is Max('Row') will return the string 'Row'. The Second is your subquery needs an alias.

Try like this:

 SELECT MAX(Row) FROM 
 (SELECT ROW_NUMBER() OVER(ORDER BY ID DESC) Row FROM USERS) UserQuery

UPDATE:

I guess there are actually 3 problems with this query :). The 3rd being, count() is a much better way as expertly described by @gbn.

• If you want to get the exact count of rows in an efficient manner, then COUNT(*) is it. The ANSI standard (look for "Scalar expressions 125") states that COUNT(*) give the row count of a table: it is intended to be optimised from the start.

If COUNT(*) is specified, then the result is the cardinality of T.

• A ROW_NUMBER() function isn't a practical option: it isn't a counting function (it's "ROW_NUMBER") and it will run badly as you add rows: a few 1000 will show how bad this it

• SUM(1) may be optimised to COUNT(*) internally but I'd never use it

• @@ROWCOUNT will require all rows to be returned by the first SELECT, which is a huge unnecessary overhead.

If you can live with approximate for SQL Server then use sys.dm_db_partition_stats. Marian's answer is out of date now since SQL Server 2005 added dmvs

SELECT
   Total_Rows= SUM(st.row_count)
FROM
   sys.dm_db_partition_stats st
WHERE
    object_name(object_id) = 'Mytable' AND (index_id < 2)

See this on SO too for some more info: http://stackoverflow.com/questions/6069237/fastest-way-to-count-exact-number-of-rows-in-a-very-large-table/6069288#6069288

In summary, there is exactly one useful way of getting the number of rows in a table. COUNT(*)

There's another variant, without scanning the table, using system tables:

Select si.RowCnt
from sys.objects so join sysindexes si on si.id = so.object_id
where so.type_desc = 'USER_TABLE'
and si.indid in (0,1) -- heap or clustered index
and so.Name in ('Users')

Compare plans, durations, time.. You'll see it's a faster variant.