Functional MRI studies on humans with BOLD contrast are increasingly performed at high static magnetic field in order to exploit the increased sensitivity. The downside of high-field fMRI using the gradient-echo echo-planar imaging (GE-EPI) method is that images are typically very strongly affected by image distortion and signal loss. It has been demonstrated at 1.5 T and 3 T that image artifacts can be reduced and functional sensitivity simultaneously increased by the use of parallel-accelerated multi-echo EPI. Using sensitivity measurements and an activation study with a cognitive Stroop task experiment (N=7) we here investigate the potential of this method at 7 T. The main findings are: (a) image quality compared to a conventional acquisition scheme is drastically improved; (b) according to CNR estimations the average BOLD sensitivity increases by 6.1+/-4.3% and 13.9+/-5.5% for unweighted and CNR-weighted echo summation, respectively; (c) both functional signal changes and sensitivity in the multi-echo data do not exhibit a pronounced dependence on TE. The consequence is that (d) in practice the performance of simple echo summation at very high field is comparable to that based on a CNR filter. Finally, (e) temporal noise observed in the different echo time courses is not strongly correlated, thus explaining why echo summation is advantageous. The results at typical spatial resolution show that multi-echo EPI acquisition leads to considerable artifact reduction and sensitivity gains, making it superior to conventional GE-EPI for fMRI at 7 T.