Brain-computer interfaces (BCIs) provide a viable alternative means of communication for persons with severe disabilities who may not have access to functional speech or gesture. BCIs based on electroencephalography (EEG) sense the electrical activity that arises from the brain via electrodes placed on top of the head. A visual BCI speller allows people to select characters of the alphabet sequentially, simply by focusing their visual attention on the desired character amidst an array of characters that are all displayed on a computer screen and illuminated rhythmically and in random order. While visual BCI spellers show great promise for providing individuals with severe disability with access to communication, they require improved speed and accuracy to be effective. To improve the speed and accuracy, we propose to detect the error-related potential (ErrP) that is elicited in participant's EEG signal when the BCI chooses the incorrect character and use this to adapt the BCI speller. We will improve accuracy by automatically correcting the erroneously selected character when ErrPs are detected. We will improve the speed, by adjusting the number of stimulus repetitions of character illumination required to choose each letter. In effect we are using the person's recognition of a BCI error to enable it to improve its accuracy.