Various inductive machine learning approaches and problem specificstious are analysed from s logical perspective. This results in a unifying framework for the logical aspects of inductive machine learning and data mining. The framework explains logicalsimilarities and differences between different machine learning settings, and allows us to relate past and present work on inductive machine learning using logic (such as structral matching, inductive logic programming, data mining, etc.). Central to the unifying framework are three dimensions: learning from entailment versus learning from interpretations, learning CNF versus DNF, learning characteristic versus dlscriminant descriptions. Though the exposition handles both first order and propositional logic, it is meant to be understandable for everyone familiar with propositional logic.