We propose in this paper a general framework for integrating inductive and case-based reasoning (CBR) techniques for diagnosis tasks. We present a set of practical integrated approaches realised between the KATE-Induction decision tree builder and the PATDEX case-based reasoning system. The integration is based on the deep understanding about the weak and strong points of each technology. This theoretical knowledge permits to specify the structural possibilities of a sound integration between the relevant components of each approach. We define different levels of integration called "cooperative", "workbench" and "seamless". They realise respectively a tight, medium and strong link between both techniques. Experimental results show the appropriateness of these integrated approaches for the treatment of noisy or unknown data.