We present a methodology for data schema integration, able to merge a set of schemas and the top-down chains of refinement planes produced during their design. The result of this process, that we call multilevel integration, is an integrated schema plus an associated top-down chain of schemas. The integrated schema and the chain are related to the input schemas by nice properties, giving rise to a two-dimensional structure that we call grid. The methodology we define is based on the formal definition of schema refinement and on the notion of schema transformation, i.e., a replacement mechanism that maps a schema into another one. In particular we concentrate on sequences of transformations, constituting the formal counterpart of a chain of refinement. The overall generation process is summarized through the notion of abstraction tree, a useful tool for discovering and solving the conflicts arising during the multilevel integration activity.