Spectrum　sensing　is　the　fundamental　technology　of　Cognitive　Radio　(CR).　This　paper　proposes　a　novel　multiple-antenna　based　spectrum　sensing　method　with　random　arrivals　of　primary　users　(PUs).　The　method　first　constructs　a　local　test　statistic　for　each　antenna　based　on　the　assumption　that　the　PU　is　randomly　arriving,　and　then　calculates　the　estimated　optimum　coefficient　for　every　local　test　statistic　and　finally　gets　the　total　test　statistic　by　linearly　combing　the　local　test　statistics　and　the　estimated　optimum　coefficients.　Given　false　alarm　probability,　the　proposed　method　calculates　the　corresponding　threshold　according　to　Neyman-Pearson　(N-P)　criteria　and　makes　the　final　decision　by　comparing　the　total　test　statistic　with　the　threshold.　Simulation　results　prove　the　validity　and　the　superiority　of　the　proposed　method　when　the　PU　is　arriving　randomly.　Simulation　results　show　that　the　proposed　method　is　superior　to　the　conventional　energy　detection　method　with　random　arrivals　of　PUs,　about　1　dB　advantage　in　performance.　©　2017　IEEE.