I have to agree with the others that adding a WHERE clause to all INSERT statements is not a good workaround, for the reasons already given (not robust, not possible for INSERT VALUES, not sufficient for multi-row insert, no protection in case of updates).

I do agree with the other four options.

It is interesting that the Microsoft interpretation of the UNIQUE constraint is not in accordance with the ANSI standard, that explicitly requires that NULL values are exempted from all constraints - so a UNIQUE constraint should not disallow multiple NULLs. And frankly, in cases where I needed a UNIQUE constraint on a nullable column, I almost always had to use one of these workarounds, which indicates that the ANSI standard is more suited for actual use.

I proposed a suggestion on Connect to gradually (to prevent breaking existing code) change the current implementation to an ANSI-standard implementation. Despite 99 votes in favor and only 4 votes against, MS has not implemented the suggested first step in SQL Server 2008 or in SQL Server 2008R2. I have now changed the proposed solution for a first step in SQL Server 2011. If you agree with me on this, then please make yourself heard, and vote for my suggestion at https://connect.microsoft.com/SQLServer/feedback/details/299229/change-unique-constraint-to-allow-multiple-null-values

Thanks in advance!