Intracranial pressure (ICP) is influenced by an array of predictable and unpredictable factors. Statistical modelling of this signal has only limited applicability because of the significant load of stochastic components. We tested the efficiency of an alternative approach, based on the methodology of artificial neural networks (ANNs) in the on-line prediction of future values of ICP and in the classification of signal properties. Satisfactory accuracy of forecasting was achieved with the ANNs for a 3-minute prediction horizon, while the prediction quality with autoregressive models of statistical origin was proved unsatisfactory. The results obtained with the ANNs were further improved when signal pre-processing with wavelet transform was employed. Nevertheless, even with the ANN methodology, no sudden breakdowns in the ICP signal (which in this respect might be compared to a "catastrophe") can be forecast with any practical applicability. We therefore applied two ANN algorithms, oriented at classification and discrimination of the global properties of the ICP signal. The neural network was expected to discriminate those sets of signal properties, which were assumed to correspond to certain clinical conditions of the patient. In a "dynamic pattern classification" the network was presented with several sections of ICP records. This was combined with information about the assignment of a given record to one of four arbitrary classes of danger. In this mode no data pre-processing was carried out, in contrast to our second approach, in which the signal was pre-processed with statistical analyses and only these intermediate coefficients were fed to the ANN classifier. The results obtained with both classification methods at their present stage of training were similar and approximated to a 70% rate of judgements consistent with expert scoring. Nevertheless, the method based on the assessment of global parameters of the ICP record seems more promising, because it leaves the possibility of extending the set of training data by information from other diagnostic modalities. The study aims towards the development of a pseudo-intelligent computer expert system, which has would be taught salient links between data extracted from the ICP signal and higher- order data, which contributed to the expert score. Hence the system would be able to make decisions on the basis of a reduced set of input information, available from a standard monitoring modality.