The minimization of cost, power consumption and development time of DSP applications requires the development of methodologies for the automatic implementation of floating point algorithms in fixed point architectures. In this paper, the influence of the DSP architecture on the computation precision is analyzed. The necessity of taking into account the DSP architecture model in the data coding process is shown. Then, a new methodology for the implementation of algorithms into fixed-point DSPs is defined.