In this talk I shall present a personal perspective on the quest for a unifying framework and theory of data mining. The starting point will be the notion of an inductive database as proposed in the seminal paper by Imielinski and Mannila (CACM 95), in which the knowledge discovery process is viewed as a querying process. The idea is that queries would return patterns and models. This framework is based on a parallel between database and data mining theory and has as ultimate goal the discovery of the equivalent of Codd's relational algebra for supporting data mining. I shall then continue to outline the more recent framework of declarative modeling for data mining, which exploits a parallel between data mining and constraint satisfaction and optimization. In this framework, data mining tasks are specified as constraint satisfaction and optimization tasks, that is, the data miner provides a model that specifies the constraints and optimization criteria that should be satisfied and a general purpose solver should compute solutions to these problems. By separating the model from the solver, a declarative approach to data mining is realized. I shall then conclude the talk by putting these frameworks into a broader perspective.