OK, thanks, much clearer now. Yep, P5NCI is what you want. Notice that P5NCI is currently limited to C functions taking up to four arguments -- which is why it took so long to build, since p5NCI.xs contains over 100,000 lines of code covering all type combinations of up to 4 arguments. The only way to avoid this brute-force approach is to manipulate the call stack with non-portable assembler, which I assume is how it's done by Win32::API and libffi (now maintained as part of GCC). See also Win32::API equivalent for Unix and Re: Disputation of g0n on the power and efficacy of XS.