An electroencephalogram (EEG)-based brain computer interface (BCI) is a novel tool that translates brain intentions into control signals. As the operational dimensions of motor imagery are limited, we describe in this paper an extension of its capability by including speech imagery. Our new system was tested with the help of subjects, whose native language is Chinese. The tests were divided into two steps. The first step was speech imagery; consequently motor imagery and speech imagery were merged in the second step. Feature vectors of EEG signals were extracted from both common spatial patterns (CSP) and cross-correlation functions; then these vectors were classified by a support vector machine (SVM). The distinguishing accuracies of two intentions were found to be between 79.33% and 88.26%. This result shows that the capability of BCI for motor imagery can be extended by combining motor imagery and speech imagery.