Brain computer interfaces (BCIs) aim to help users send commands to the external world directly while bypassing the brain's normal output pathways of peripheral nerves and muscles, which is particularly important for patients whose normal means of interaction are compromised by central nervous system damage. The performance of current BCI is not precise and natural enough to be applied in human daily life and more efforts are needed to obtain the ultimate goal. In this thesis, we are trying to develop online BCI application, during which ongoing brain signals are analyzed and commands are detected and issued from time to time, to test one of human perception, color stimulus, and improve the performance of BCI offline, during which analysis are carried out after each trial, in the system and brain signal parts. Color is a fundamental aspect of human perception and could be natural for human to be used in BCI control. An online BCI game based on decoding of users' attention to color stimulus was developed to control external world using the brain signal directly. Our results showed that an accuracy of 70%-80% could be achieved and it provided evidence for the possibility in applying color stimuli to BCI applications. Currently there are two kinds of BCI systems. Fully asynchronous BCI systems suffer high signal detection problem while in synchronous systems users have to issue commands all the time. A semi-self-paced BCI in which subjects are externally paced but allowed to determine on which prompting cycles control commands was proposed. A single task with randomly interleaved trials was designed to test whether it can be used as stimulus for inducing initiation and non-initiation states. Further, the essential problem on the discrimination between initiation state and non-initiation state was studied. The final results demonstrate the viability of our proposed idea for a BCI design based on conventional EEG band powers. In the brain signal side, the accuracy of predicting movement parameters were improved by considering EEG activity during the movement planning phase. We proved that EEG activity from the motor planning phase could predict the peak speed and acceleration with a reasonable accuracy. Moreover the decoding accuracy of the peak speed and acceleration could both be improved by combining band powers from the motor planning phase with the band powers from the movement execution. In addition, reaction time and endpoint error were correlated to alpha and beta band power changes during planning reaching movements. The objective of our work was to test an online BCI game based on color stimulus and to improve the BCI performance in brain and system aspects as much as possible. More patients will benefit from the BCI if more kinds of stimuli are explored. Our work showed that color stimulus could be one alternative stimulus applied in online BCI application. Moreover, based on our proposed semi-self-paced BCI system and offline results of predicting movement parameters using band powers from the motor planning phase and the band powers from the movement execution, future BCI applications would be more practical and efficient.