We propose a method of test selection based on formal specifications, for functional testing of object-oriented software. This method is based on rigorous theoretical foundations. To limit the size of test sets, we present several hypotheses specific to object-oriented systems, used to reduce an exhaustive test set into a pertinent test set while preserving its qualities: regularity hypotheses are used to constrain the shape of tests while uniformities, with or without subdomain decomposition, are used to limit the values of the variables appearing in the tests. Care is taken to provide a method for which operational methods can be implemented.