The task of translating data from one schema into another is usually performed with the help of information stating how the elements between two schemas correspond. Translation mechanisms can use this information in order to identify how instances of a source schema must be translated. We claim that a uniform matching approach, where instances of a source classes are always translated into the same target classes, may not represent the reality, specially when the schemas involved describe taxonomies. In this paper we demonstrate taxonomies that support our idea, and propose the usage of a conditional matching approach to improve the accuracy of taxonomical instances translation.