This thesis puts together the statistical techniques indicated for modelling the mean and dispersion of quality characteristics of products and processes via unreplicated factorial experiments, resulting in the definition of an integrated and detailed script for the analysis. It was motivated by the fact that, although there are many publications about classic linear regression, generalized linear models (GLMs), response transformation and design of experiments, there is no one text which put together and describe in detail all the aspects of the modelling of the mean and the dispersion in factorial experiments. The few texts on the subject do not describe a number of important aspects in studies of this nature, e.g. how significance tests for the coefficients in GLMs should be applied and which are the statistics and plots indicated for checking model adequacy. In addition, nothing was found in the literature about model identification in factorial experiments. All these aspects are detailed in this thesis. Once the model is built, we show how to use it in order to obtain the optimal operating conditions for products and processes. Besides achieving this main objective, the thesis brings some additional contributions, namely: a) it points out limitations in all the four methods in the literature which have the purpose of selecting the most adequate transformation of the response; b) it proposes using response transformation methods as a source of indication of the link function to use in GLMs, and c) it proposes using the loglikelihood function for the joint choice of the probability distribution and of the link function in GLMs.