Differential frequency hopping (DFH) was believed to be an effective FH technique for increasing data rate of FH communication system. Cognitive radio (CR) users can use the CR technology to sense the surrounding RF envi ronment, search for available spectrum resources and ac cess spectrum dynamically. The cognitive radio technol ogy and applied mathematics are introduced in DFH. And the Latin squares based frequency set adaptive optimiza tion (LS-FSAO) algorithm is proposed. In short, the fre quency set is divided into many frequency micro-sets after the spectrum sensing, and the micro-sets are assigned to DFH users with dynamic Latin squares. The users' fre quency micro-sets are dissimilar during a same time slot to cut down the Multiple Access Interference (MAl), Par tial Band Jamming (PBJ) and Multi-Tone Jamming (MTJ). The scenario simulations of the LS-FSAO based DFH network and the traditional DFH are performed, respec tively. The results show that the LS-FSAO DFH can ob tain favorable network performance.