The recognition of Business documents like forms, invoices, postal mail,  is one of the most important application in  Document Image Analysis (DIA) which concern many private companies and public organizations in the world. Business documents processing is also a challenging problem, rarely studied in the scientific literature. These documents are complex and have heterogeneous contents. Business documents contain degraded printings and handwritings which overlapped complex coloured backgrounds or pre-printings forms. Moreover, there is no unique model of business document, each private company or public organization, design their own template. Consequently, the layout and the logical structure vary from a company to an another. Numerous commercial systems exist, but they simplify the recognition process by using a manual model which designs the informative zones for each type of document. Most of existing software use a large database of predefined models which describes the template of each existing business document from the main companies. This solution is suited for large companies which process the same model of documents from other large companies. For small companies which have to process daily a large quantity of documents from different origins, the manual modelisation is impossible to achieve. The only solution consists to develop a system which recognizes the  logical structure without any predefined model. A data-driven recognition must replace the model-driven classical approach.  We will present the main works from the state of the art in image pre-processing and segmentation, layout analysis and logical structure recognition applied to business documents.