It is well known that recording electrical signals from the human body is a challenging task due to their nature and its vulnerability to noise effects. This is also the case of the electrogastrographic signal, the electrical activity of the digestive system that can be recorded using surface electrodes on the external wall of the abdomen. This signal is especially difficult to record due to the different signals present in the abdomen that are larger in amplitude and sometimes completely mask the signal of interest. Therefore, it is very important to clean the recordings before any information can be extracted. This paper describes a methodology to segment the signals with useful information, eliminating those segments masked with artefacts that make the signal useless. A database with 50 recordings from healthy volunteers and diabetes type II patients were analyzed during three stages: fasting, after drinking water and after eating a caloric meal. The recordings are first filtered with an adaptive high-pass filter, then it is analyzed in one second windows to look for attenuations, saturation and drastic changes. A file with labels is generated that contains only the information of useful windows in order to extract the 2.5 minutes segments that provide information. Then, the Discrete Wavelet Transform and the Fast Fourier Transform are used to calculate the percentage of prevalence in the gastric pacemaker (normagastry, taquygastry and bradygastry). This automatic procedure was tested comparing the segmentation made by an expert by visual inspection. It was found that the automatic segmentation, made by the proposed algorithms, recovers 20% more segments than the expert and saves considerable the time spent in the labelling. The prevalence obtained after segmentation is congruent with the previous works reported.