Abstract: We present an approach for proving behavioral properties of numerical programs by analyzing their compiled assembly code. We focus on the issues and traps that may arise on floating-point computations. Direct analysis of the assembly code allows to accurately take into account architecture- or compiler-dependent features such as the possible use of extended precision registers. The approach is implemented on top of the generic \Why platform for deductive verification, which allows us to perform experiments where proofs are discharged by combining several back-end automatic provers.