This letter concentrates on the spectrum sensing (SS) problem for noncircular (NC) signal in multiantenna cognitive radio networks. A novel detection method called NC local average variance (NC-LAV) is developed to exploit the NC characteristic of the primary signals for SS. The distribution of the NC-LAV statistic in the absence of primary user is first derived. Then, based on the derived distribution, we present an asymptotic threshold determination of the proposed NC-LAV method. Compared to the peer methods, the proposed NC-LAV method exhibits much lower complexity. Meanwhile, the simulation results reveal that the proposed method achieves better performance than several state-of-the-art methods.