Most of the time, the mathematical formula is directly used in the Scilab source code. But, in many algorithms, some additional work is performed, which takes into account the fact that the computer does not process mathe- matical real values, but performs computations with their floating point rep- resentation. The goal of this article is to show that, in many situations, Scilab is not naive and use algorithms which have been specifically tailored for float- ing point computers. We analyze in this article the particular case of the quadratic equation, the complex division and the numerical derivatives. In each example, we show that the naive algorithm is not sufficiently accurate, while Scilab implementation is much more robust.