Hi Bronek, It seems from Jonathan's and your info the overloads uBLAS needs (for float types) should be in cmath. As you show the work around is to #include <cstdlib> I would like to make this library version dependant but for now have simply place the include in uBLAS's "config.hpp" header. Looking at the regression tests there is a far more serious problem. There is a series of warning thus: ublas/matrix.hpp", line 1030: warning #504-D: nonstandard form for taking the address of a member function size1_ (ae ().size1 ()), size2_ (ae ().size2 ()), data_ (functor_type::size1 (size1, size2) + 1) { The warning point to the functor_type::size1(size1,size2). The compiler is getting this very wrong. There is no address to be taken here! It should be calling the static 'size1' member of the functor_type. This is probably the cause of a few further failures. Michael