[en] This paper presents a number of experimental data processing notions with the aim of developing an NDT method based on merging ultrasonic and gamma radiographic data. We first review the industrial context concerned and, before moving on to specific data merging problems, we discuss the difficulties of reconstruction using only one type of data (radiographic or ultrasonic). The main part of the report begins with a brief reminder of gamma radiation and ultrasonic wave propagation principles. Certain imaging and reconstruction methods conventionally used for each type of measurement are also presented. Reconstruction problems are then directly approached in algebraic form. For the type of problem studied, the inspection data alone cannot lead to satisfactory reconstructions and we evidence the need to regulate the problem by introducing deductive information on the object to be reconstructed. The Bayes' approach provides a self-consistent means of integrating both the data information and the deductive information. It is based on probabilistic models of the variables involved, notably those of the object sought. We discuss at some length certain models of images used in gamma radiography (independent variable fields, variables having a Markov-type structure) and the Bernoulli-Gauss-type models used for ultrasonic trace deconvolution. Finally, we outline data merging paths. A formal Bayes' framework is used to present two merging approaches, after which we briefly describe our projects for the processing of already available experimental data. (author)