In recent years, steady-state visual evoked potential (SSVEP) based brain-computer interface (BCI) has received more and more attention in the research of BCI systems due to its three advantages: good system performance, little user training and ease system configuration. However, there is much room for further improvement in current SSVEP-based BCI, such as solving the problems of harmonic frequency and insufficient flickers as well as boosting its accuracy. For these reasons, we propose an improved visual stimulus generation method and to embed the phase information in the traditional feature detection algorithm. The former method is able to 1) generate the stimulus phases with a very high resolution of 360/nV (deg) while employing a monitor as visual stimulator, which overcome the problem of insufficient stimulus phases, 2) generate the suitable flickers’ phases for evoking SSVEPs at harmonic frequencies with different phases which help overcome harmonic frequency problem. The latter one may boost the accuracy of SSVEP-based BCI beyond 10% against the traditional one. Furthermore, although the proposed feature detection algorithm only utilizes the single signal electrode, it can overcome harmonic frequency problem and achieve the improved accuracy as similar as the multi-channel technologies: canonical correlation analysis (CCA) and minimum energy combination (MEC). Finally, the first online SSVEP-based BCI system based on both frequency and phase information is implemented. It may achieve the average online accuracy rate of iv 94.81% and ITR of 38.64 bits/min respectively in despite of employing the harmonic frequencies of 17.14, 8.57, 15 and 7.5 Hz and only one signal electrode. In addition, this performance can be comparable to the other analogous systems.