Tuning a fuzzy system to meet a given set of requirements is usually a difficult task that involves many parameters. Since doing it manually is often cumbersome, several CAD tools have been reported to automate this process. The tool we have developed, xfsl, tries to reduce the limitations of other tools. In this sense, it includes a wide set of supervised learning algorithms and is able to cope with complex fuzzy systems. In particular, xfsl is able to adjust hierarchical fuzzy systems; systems that employ fuzzy functions defined freely by the user, like membership or connective functions, defuzzification methods, or even linguistic hedges; and fuzzy systems with continuous outputs (such as fuzzy controllers) as well as categorical outputs (such as fuzzy classifiers). Several examples included in this paper illustrate all these issues. Another relevant advantage is that xfsl is integrated into the fuzzy system development environment Xfuzzy 3, and, hence, it can be easily employed within the design flow of a fuzzy system. © 2007 Elsevier B.V. All rights reserved.