Methods for obtaining a stable solution in the case of improperly posed problems of mathematical physics have, recently, undergone considerable development; the applications of these methods to particle physics have been widely discussed (see, for instance, ref. (i) and the papers quoted there). More precisely we recall that, in many cases, it is necessary to perform a numerical analytic continuation of the scattering data in order to extract the information that one is looking for. In these cases one must take into account two opposite aspects of the problem; the analyticity of the functions and the noise of the experimental data. Moreover let us observe that to extract information through the numerical analytic continuation of the scattering data can be regarded as an inverse problem in a rather generalized sense. ~n the present note, following TIKttONOV and ref. (~), we call direct problems those which are oriented along a cause-effect sequence (i.e. problems of finding out the consequences of given causes) ; in this sense the inverse problems are those associated with the reversal of the chain of causally related effects, i.e. problems of finding the unknown causes of known consequences. In other words, if we assume the previous definition, not only the reconstruction of the potential from the knowledge of the angular distr ibution (nonrelativistic inverse problems at fixed energy) but also the determination of a coupling constant or of the spectral function in a dispersion relation and so on are inverse problems in a generalized sense. From this point of view these problems look very similar to the reconstruction of the input signal from the reaction at the output of a device. In this sense, some of the S-matrix methods used in particle physics appear to be a particular case of the more general methods of system theory. Nor this is surprising, since, as is well known, the dispersion relation approach was originally applied in electrical network analysis and in optics and successively used in particle physics. Now many inverse problems are improperly posed (see, for instance, ref. (3)). The more appropriate way of approaching these ill-posed problems seems to be a probabilistic method, which should take into aeemmt the random nature of the noise. To this