Brain-Computer Interfaces (BCI) are used to provide environmental interaction among individuals, especially people with disabilities. Spelling systems, one of the BCI applications, are based on the principle of detecting P300 waves from EEG signals. The aim of speller systems is to identify the P300 waves and determine the letter on a screen opposite the person. The purpose of the operating speller systems is to minimize the processing cost of the system with smaller data sizes to be obtained by compressing EEG data before the pre-processing step. In this study, a hybrid model was presented. With Pairwise and variance-based signal compression Algorithm, first of all, data is compressed and then preprocessing, and classification is performed. The proposed hybrid model is intended to be stored in offline systems and to increase the speed of operation in online systems.In this paper, we proposed a new hybrid model with the compression algorithm called Pairwise and Variance Based Signal Compression Algorithm (PVBSC) for P300-based speller systems. The proposed method wasevaluated on Wadsworth BCI speller dataset. As the focus is the compression algorithm, the channel selection has been applied to increase the working speed. Channel selection was made by detecting eight channels most commonly used in the literature.As the first step in the compression algorithm was segmentation, the study was repeated with 16, 32 and 64 channel lengths to see the effect of the window length. Then, to find the target character from EEG signals, we have used two different classifiers including an ensemble of LS-SVMs and ensemble of LDAs. In this study, as the best classification accuracy, 1.437 compression ratio and 94.166% accuracy rate by Ensemble of LDAs was achieved with PVBSC with 32 window lengths.The obtained results have shown that the proposed compression method could be confidently used in the compressing the P300 wave-containing EEG signals and reduce the data size significantly.