counting the number of rows in a table

If you can deal with a rough count on SQL 2000 - you can query sysindexes - something like

SELECT rowcnt FROM sysindexes WHERE id = OBJECT_ID('MyTableName') and indid in (0, 1)

Under SQL 2005+ the equivalent would be

SELECT [record_count] FROM [sys].[dm_db_index_physical_stats](DB_ID(), OBJECT_ID('MyTableName'), NULL, NULL, 'SAMPLED')
WHERE [index_id] IN (0, 1)

The row count returned from 2005+ will be accurate, unless your table has no clustered index (i.e. is a heap).

These methods are orders of magnitude faster than counting all the rows using COUNT().

EXEC sp_spaceused 'MyTableName'

one of the returned columns is "rows"

Alternatively, of course, you can run:

SELECT COUNT(*) FROM MyTableName

If you need to store the count in a variable, then use this:

SET @MyVar = (SELECT COUNT(*) FROM MyTableName)

I use the dm_db_index_physical_stats as well. it runs nightly as part of maintenance

You could also use:

SELECT row_count 
FROM sys.dm_db_partition_stats
WHERE OBJECT_ID = OBJECT_ID('TableName')
    	AND index_id =1

I specified index_id = 1, since you mentioned that all table have a PK, which I inferred to be the clustered index. You could also specify index_id < 2, which would account for a HEAP table