The complete loss of motor skills caused by the suffered traumatic event or as a result of neurodegenerative disease such as amyotrophic lateral sclerosis, almost always manifests itself through inability to communicate, both in writing and speech, which further degrades the life quality of the individuals in question. In case of a patient’s functional eyesight, the assistive technology offered a whole series of technological solutions which restore the communication capability of the mentioned population, increasing their life quality. Despite good results and a wide selection of options, such systems are not suitable for a small group of people who experienced vision loss in the advanced stages of diseases. Problem of communicating with such individuals remains the open topic, where current solutions provide very limited results. A potential solution is available through the application of a brain-computer interface, i.e. a system which converts the measured and processed neurological activity into control commands for the managed system. Large number of such interfaces use a neurological reaction to visual stimuli or use some form of visual support in order to simplify the control of the system, hence this type of interface is also limited to individuals with eyesight. This doctoral thesis proposes brain-computer interface for writing that applies only auditory stimuli, and which is intended for the mentioned group whose neurological activity is measured by the most widespread and affordable method: Electroencephalogram (EEG). The basic idea behind the proposed system is that the user selects the desired letter from a series of randomly ordered letters of the Croatian alphabet presented through the auditory system based on the classified measured brain activity. The selection of stimuli in the form of letters, that is, phonemes from the Croatian language enabled the implementation of a spelling system maximally shortening the end-user training time. Processing and analysis of neurological reactions measured on a group of subjects with the proposed interface showed stable typical brain reactions to target stimuli, and the same data has been used to train several types of classifiers and spatial filters. Validation of the system has been performed on the same group of subjects through the task of writing sentences, showing a satisfactory result in 50% of subjects. The final system implementation includes processing and classification of measured EEG signals in real time across two classification levels. In the first level, all included letters are presented to the subject once, while the second classification level further increases the classification accuracy by stimulating the subject with additional five letters which achieved the highest classification probability in the previous stage. The limitations of the proposed EEG measurement and processing method is shown in the analysis of the system accuracy with the reduced number of used electrodes. The obtained results indicate a similar system accuracy with the reduced number of electrodes, especially by subjects who are able to use proposed system with great confidence, indicating a deficiency in the spatial resolution of the EEG measurement which can be solved by using measuring methods with better spatial characteristics.