>It is not a fixed rule.
>In some cases, the engine realizes the indexes and generates a good plan, in practice flowing the two indices and makes a union at the end.
>On next example, the simplicity allows the engine to generate a good plan,
>but if you compare the plans you see that the one with union is simpler.
>
>

>DROP table #TT 
>create table #TT (AA INT PRIMARY KEY, BB INT)
>CREATE INDEX IK ON #TT (BB)
>
>INSERT #TT	SELECT	2*ROW_NUMBER() OVER (ORDER BY (SELECT NULL))
>				,	3*ROW_NUMBER() OVER (ORDER BY (SELECT NULL))
>				FROM sys.objects A,sys.objects B
>
>SELECT * FROM #TT WHERE AA = 4 OR BB = 3000
>
>SELECT * FROM #TT WHERE AA = 4
>	UNION
>	SELECT * FROM #TT WHERE BB = 3000
>
>SELECT * FROM #TT WHERE AA = 4 OR BB = 6
>
>SELECT * FROM #TT WHERE AA = 4
>	UNION
>	SELECT * FROM #TT WHERE BB = 6
>

I tried all that. It is interesting to see the adjustment in the execution when using UNION.

Thanks