Grid data provides finely localized information allowing a user to analyze a posteriori, the zoning that would be specific to him by grouping cells. An important issue with this kind of data is the confidentiality; the smaller the cell, the greater the risk of statistical disclosure.

This abstract briefly presents two methods : one based on a "quadtree" type algorithm allows to build geographical areas where the primary secret is respected, the other formalizes the secondary disclosure issue with a graph representation and allows a quick detection of areas where disclosure by geographical differentiation is possible.