A brain-computer interface (BCI) provides a direct communication channel between a brain and an external device. Steady-state visual evoked potential based BCI (SSVEP-BCI) has received increasing attention due to its high information transfer rate, which is accomplished by individual calibration for frequency recognition. Task-related component analysis (TRCA) is a recent and state-of-the-art method for individually calibrated SSVEP-BCIs. However, in TRCA, the spatial filter learned from each stimulus may be redundant and temporal information is not fully utilized. To address this issue, this paper proposes a novel method, i.e., task-discriminant component analysis (TDCA), to further improve the performance of individually calibrated SSVEP-BCI. The performance of TDCA was evaluated by two publicly available benchmark datasets, and the results demonstrated that TDCA outperformed ensemble TRCA and other competing methods by a significant margin. An offline and online experiment testing 12 subjects further validated the effectiveness of TDCA. The present study provides a new perspective for designing decoding methods in individually calibrated SSVEP-BCI and presents insight for its implementation in high-speed brain speller applications.