To investigate the directionality of neural interactions as assessed by electrophysiology, we adapted methods of structural analysis from the field of econometrics. In particular, within the framework of autoregressive modelling of the data, we considered quantitative measures of linear relationship between multiple time series adopting the Wiener-Granger concept of causality. The techniques were evaluated with local field potential measurements from the cat visual system. Here, several issues had to be addressed. First, out of several statistical tests of the stationarity of local field potentials considered, those based on the Kolmogorov-Smirnov and on the reverse arrangement statistics proved to be most powerful. The application of those tests to the experimental data showed that the large part of the local field potentials can be considered stationary on a time scale of 1 s. Second, out of the several investigated methods for the determination of an optimal order of the autoregressive model, the Akaike Information Criterion had the most suitable properties. The identified order of the model, across different repetitions of the trials, was consistently 5-8. Third, although the individual segments of field potentials used for the analysis were relatively short, the methods of structural analysis applied produced reliable results, confirming findings of simulations of data with similar properties. Furthermore the features of the estimated models were consistent among trials, so that the analysis of average measures of interaction appears to be a viable approach to investigate the relationship between the recording sites. In summary, the statistical methods considered have proved to be suitable for the study of the directionality of neuronal interactions.