Thanks to their ability to model natural language, fuzzy rules are very popular in expert knowledge representation. Mamdani fuzzy systems are widely used for process simulation or control. Nevertheless, fuzzy implicative rules, and especially gradual rules, provide another kind of knowledge representation, which can be very useful in approximate reasoning. In this paper, the two types of rules are compared according to their behavior in some typical situations such as rule interpolation, combination of a specific rule with a more generic one. The comparison is carried out with regard to the output possibility distribution, the crisp inferred value and the rule base consistency. Finally, we discuss the complemental aspects of these rules and we show how in certain cases gradual rules may constitute an interesting alternative to Mamdani rules.