Steady-state visual evoked potential (SSVEP)based brain-computer interface (BCI) speller has the potential to carry out rapid communication between the human brain and the external device. In SSVEP-based BCI spellers, each target character is blinked with a unique frequency depending on the target; in other words, the unique visual stimulus is assigned to each target character; thus, attention of the targeted character may generate a unique SSVEP depending on visual stimulus. However, most SSVEP-based BCI spellers have adopted a cue-guided target approach that guides both resting (stop) and task (go) by a predefined external cue. In this paper, we proposed a method to detect the user’s cognitive state that provides working state information from electroencephalogram (EEG) signals in the SSVEP-based BCI speller. The method proposed consists of two steps: one procedure estimates the difference between two frequency band powers (narrow band and interval band stimulus frequency), and the other classifies cognitive states using filter bank canonical correlation analysis (FB-CCA). To investigate our proposed method’s feasibility, we recruited 40 participants to perform the 40-class SSVEP-based BCI speller. Our proposed method achieved 71.7± 25.2% classification accuracy without an external cue, while the conventional cue-based method achieved 73.1± 22.2%. Thus, our proposed and cue-based methods yielded no significant difference in performance (student’s paired t-test, p=0.41), showing that users may operate the SSVEP-based BCI speller without an external cue with no loss of performance.