In this study, we propose an improved semi-supervised support vector machine (SVM) based translation algorithm for brain-computer interface (BCI) systems, aiming at reducing the time-consuming training process and enhancing the adaptability of BCI systems. In this algorithm, we apply a semi-supervised SVM, which builds a SVM classifier based on small amounts of labeled data and large amounts of unlabeled data, to translating the features extracted from the electrical recordings of brain into control signals. For reducing the time to train the semi-supervised SVM, we improve it by introducing an batch-mode incremental training method, which also can be used to enhance the adaptability of online BCI systems. The off-line data analysis results demonstrated the effectiveness of our algorithm.