In general, paradigms for fMRI experiments consist of one or several task parts and a resting state part. Activations are detected for each voxel statistically by taking the resting state as a reference. This conventional method can be inefficient for the experiments with multiple stimuli. We present a new fMRI data analysis method which can detect activations without imaging the resting state references, but by statistical estimation of the reference from the acquired signal using the multiple stimuli. The experimental results are reliable to each stimulus even in case there are overlaps of their activation regions. Then, the experiment time is saved due to eliminating the resting state imaging.