This paper introduces a novel approach to study the quaternion‐valued memristive neural networks and uses this method to investigate the Lagrange stability of the memristive system. By utilizing the logical switching signal, a stop and go strategy is proposed, which makes the model translated into a system contain two different switching parameters; thus, the derived conclusions contain much more information of the states. Moreover, the difficulty of comparing the “size” of quaternions can be avoided. In the case of discrete‐time system, a very useful lemma is presented, which makes the analysis of the discrete‐time quaternion‐valued memristive network easy to operate. Through simulations, the main results regarding to the Lagrange stability are illustrated.