The present paper argues that generalisation is conservative. Our goal was to experimentally study the links between knowledge generalisation and the storage of contextual elements. The knowledge domain, very simple chess configurations, allowed subjects, novices in chess, to acquire micro-expertise based on the analysis of a single source problem. In the first experimental phase, subjects had to analyse a source problem. We induced two modes of source-problem encoding: In the first group, subjects were given explanations focused on the sequence of elementary solving steps; in the other group they were given the general principle relevant to the category of problems in question. Subjects had then to solve different tests (solving isomorphic problems, recall tests, similarity tests) designed to answer two questions: The first question was to test whether the experimental manipulation in the two groups had in fact generated knowledge that varied in abstractness; the second question was to determine whether generalisation is accompanied by storage of surface features of the source problem. Results show that the knowledge generalisation is conservative. Subjects who generalise their knowledge have a better memory retention of context-dependent elements than the other subjects.