In this paper, we introduce a spectrum holes detection approach, which combines blind modulation classification, angle of arrival estimation and number of sources detection. We perform eigenspace analysis to determine the number of sources, and estimate their angles of arrival. In addition, we classify detected sources as primary or secondary users with their distinct second-order/one-conjugate cyclostationarity features. Extensive simulations carried out indicate that the proposed system identifies and locates individual sources correctly, even at -4 dB SNR.