Sensitivity analysis can be used to quantify the magnitude of the dependency of model predictions on certain modelling assumptions, e.g. parameter values, initial conditions or inputs. The finite difference method, a local sensitivity analysis technique, is discussed in detail and situated among other methods. A lot of attention is paid to the practical issues concerning the implementation of this technique, more specifically the effect of nonlinearities of the model and numerical problems. The influence of the perturbation factor on the sensitivity calculations is investigated and different criteria are proposed to assess the quality of the sensitivity functions. A threshold value with good probability of detecting faulty sensitivity function calculations was found for one of these criteria, implying that the method can be automated.