The goal of this work was the gathering of enough information about training techniques for fuzzy systems in order to be able to choose the best one to be used with specific input/output data pairs. For this, documentation referring to training methods was reviewed, implementing them under the MATLAB© environment. This with the objective of evaluating results, and including the tests involving clustering. Analysis of the test results showed that the tools implemented gave better results than those given in the specialized literature, with the exception of methods of theoretical descent, from which the results obtained were not those expected. The least-squares methods and those of backpropagation gave good results for deterministic universes.