With the large expansion of available textual data, text mining has become of specialinterest. Due to their unstructured nature, such data require important preprocessing steps.Among them, stemming is a popularly used preprocessing method that extracts the root of thewords. However, the most popular algorithms are based on the application of rules, and there-fore highly language-related. We propose a new approach, the RFreeStem, that is rather basedon corpus and can therefore be applied on many languages.