Target detection is one of the most important aspects in remote sensing theory and application. Hyperspectral image can provide radiation,geometrical and spectral information of targets simultaneously,making target detection much better than other methods. A target detection algorithm based on variance minimum (BVM) which makes use of highlighting information of detection results is presented. And two experiments on different spatial resolution and spectral resolution are conducted to compare BVM method and constrained energy minimization (CEM). Results show the more robust performance of BVM method.