We present the construction of the Experimental Probabilistic Hypersurface (EPH) which introduces probabilistic methods in order to reconstruct or predict some information resting upon the data already given or calculated. The feature of this method is its relying upon a general principle of maximal entropy, thereby we do not place any kind of artificial assumptions as a basis. Originally the EPH was constructed in order to meet a request from Framatome ANP. Consequently EPH was significantly advanced and improved following the contracts with the IRSN. It was dealing with the computation code named «CATHARE» which calculated the temperature reached in a nuclear reactor in case of an accident. A few hundred runs were performed and the concern of EPH was to evaluate the probability to overstep a given threshold in «unexplored» part of the space. Along with the analysis of uncertainties in a uniform medium, EPH handles the propagation of information in a non homogeneous space. Another application of the EPH is devoted to a problem coming from the European Environment Agency: for a given concentration of a pollutant, measured by many stations for several years, we need to reconstruct the missing years and predict the future evolution.