In the Estimation of Distribution Algorithm (EDA), the distribution of the high performance solutions is presented by a so called probability model. From this point of view, a memory enhanced estimation of distribution algorithm (M-EDA) is proposed to solve the dynamic optimization problems (DOPs). In the M-EDA, a probability model is stored as the basic memory element and reuse under the new environments. A memory management scheme combines the best individual method with the samples averaging method is designed and the population diversity is compensated dynamically. The experiment results show the universal property of the M-EDA and verify the ability of the diversity compensation methods to maintain the diversity of the population. In the experiments on five dynamic optimization problems, M-EDA performs significantly better than the other two state-of-art dynamic evolutionary algorithms.