Аннотация:Methods of the linear theory of computer-aided measuring systems are well developed. They allow obtaining the most accurate estimates of the parameters of the object under study from the measurement data (the reduction of measurement), as well as monitoring the consistency of the used mathematical model with the measurement result. In this paper, these methods are generalized to a class of nonlinear estimates implemented using neural networks. Sample estimates of the accuracy of the reduction of measurements and the agreement of the model with the data are used. The approach is applied for estimating atmospheric parameters based on spectral measurements of scattered solar radiation.