The problems of modelling the yield of crops cultivated in unfavourable natural conditions of risky farming under excessive heat supply and deficit of natural precipitation are considered. Time series (TS) of crop yields on the example of grain crops grown in the Lower Volga region of Russia were used as input data for modelling. Built-in tools of the integrated mathematical and statistical environment Statistica v.10 were used for computer modelling. Preliminary analysis was carried out using spectral analysis based on Fourier series. In the process of modelling, a family of artificial neural networks of multilayer perceptron architecture was built, the parameters of which were selected in an automated mode. The quality assessment of neural networks was checked by the MAE criterion on a test sample. The obtained results confirmed the acceptability of using the developed neural networks for solving the problem of computer modelling of crop yields of grain crops grown in arid conditions under moisture deficit.