In this paper, a class of fractional-order octonion-valued recurrent neural networks (FOOVRNNs) with impulsive effects and delays is discussed. Firstly, considering the multiplication of octonion numbers does not satisfy the commutativity and associativity, we separate the octonion-valued system into four complex-valued systems. Secondly, we obtain the global asymptotical synchronization of FOOVRNNs by applying the appropriate Lyapunov function. Finally, we give two illustrative examples to illustrate the feasibility of proposed method.