The embodiment of the invention provides a virtual machine management method and device, and relates to the technical field of communication. The repetitive oscillation probability of a network element between capacity expansion operation and capacity reduction operation can be reduced. The method comprises the following steps of: predicting a first theoretical load value after a virtual machine unit of a target network element is expanded from current M VM to M+X VM, wherein X is predetermined capacity expansion step length; X is greater than or equal to 1; and M is greater than or equal to 1; according to the first theoretical load value, determining a first capacity reduction threshold value of the target network element, wherein the first capacity reduction threshold value is less than or equal to the first theoretical load value; and, if the current load value of the target network element is less than the first capacity reduction threshold value, releasing Y VM from the virtual machine unit of the target network element according to the predetermined capacity reduction step length, wherein Y is greater than or equal to 1.