Objective. Idle mode detection is a vital problem to be solved in self-paced asynchronized BCI, because patients need to control the BCI system whenever he or she wants, rather than output commands according to the system cues which is essentially different between self-paced and synchronized BCIs. With the detection of idle mode, we can finally increase the performance of real-time BCI. Approach. In this work, we introduce a new experiment paradigm to research the difference between idle mode and task mode from those domains of time, frequency and spatial. The experiment is carried out with 14 volunteers from 19 to 26 years old. Main results. Off-line analysis shows significant differences of the distribution of EEG power spectrum between task mode and idle mode. When the subjects execute a left or right hand selective sensation, there appears an obvious ERD/ERS in the subject's contralateral cortex. However, there is no ERD/ERS during idle mode periods even with simultaneously applied vibration. With combining data set of left and right hand sensation for classifier calibration, we recognize idle mode from mixed task mode with TPR value of 86%, and this result is significantly higher than that TPR achieved from single task type left or right hand sensation with $$p<0.05$$. Significance. The new proposed calibration method is demonstrated to be feasible for idle mode detection in SS-BCI. With the combination of two mental tasks, we can improve the BCI performance by increasing true positive rate, and limit false positive rate at the same level compared to the idle mode recognition from single task type.