The great effort needed for manual schema matching, in data migration, data warehousing and real-time query translating, has induced some researchers to look forward for an automatic mapping procedure among heterogeneous databases. These methodologies use rule-based systems, fuzzy-logic, mathematical calculations, algorithmic solutions and also Artificial Neural Networks (ANN). In the case of ANN, especially a tool called SemInt developed by Li and Cliffton, which performs a one-to-one mapping, can be considered as a pioneer work in this area. In this paper, our goal is to propose an alternative methodology to SemInt, on a Backpropagation ANN basis, but using a representing schema, which allows a many-to-many cardinality, a feasible alternative in case of well-known and stable domains. Data pre-process, relevant input definitions, and a sample reference schema are considered. Topologic-level concept is introduced and its application showed by examples from real practice. A backpropagation neural network is developed, and trained following the methodology described in this work.