Thanks Matt for the comparisons. My experience has been similar with floating point calculations, between 50% to 90% improvement. For text (regex-type functions), the improvements were many fold. Of course the comparable T-SQL functions could themselves be improved but for the quickest bang-for-the-buck, CLRs really shine when the functions are well-defined. In a couple of cases, I took some previously existing functions written in C++ and made them into CLR functions so the calculations came out identical to external apps. This avoided a lot of discrepancies between T-SQL and external apps.