A dendritic description of multilayer networks, with Radial Basis Units, is applied to a real classification problem of financial data. Simulations demonstrate that the dendritic description of networks is suited for classification where input data is divided in subspaces of similar information content. The input subspaces with reduced dimensions are processed separately in the hidden stages of the network and combined by an associative stage in the output. This strategy allows the network to process any combination of the input subspaces, even with partial data patterns. The division of data also permits to deal with many input components by generating a set of data subspaces whose dimensions have a manageable size.