We propose a novel method for the extraction of discriminative features in electroencephalography (EEG) evoked potential latency. Based on our offline results, we present evidence indicating that a full surround sound auditory brain–computer interface (BCI) paradigm has potential for an online application. The auditory spatial BCI concept is based on an eight-directional audio stimuli delivery technique, developed by our group, which employs a loudspeaker array in an octagonal horizontal plane. The stimuli presented to the subjects vary in frequency and timbre. To capture brain responses, we utilize an eight-channel EEG system. We propose a methodology for finding and optimizing evoked response latencies in the P300 range in order later to classify them correctly and to elucidate the subject’s chosen targets or ignored non-targets. To accomplish the above, we propose an approach based on an analysis of variance for feature selection. Finally, we identify the subjects’ intended commands with a Naive Bayesian classifier for sorting the final responses. The results obtained with ten subjects in offline BCI experiments support our research hypothesis by providing higher classification results and an improved information transfer rate compared with state-of-the-art solutions.