Reservoir computing approaches have been around for almost two decades. They were developed to solve the difficult gradient-descent training of recurrent neural networks. However, in reservoir computing, the choice of parameters and architecture often leads to an optimization challenge. Most early applications have used trial and error with expert knowledge to select the right value for parameter(s)/architecture. This approach is usually cumbersome and difficult considering the large search space. Metaheuristics have been known to perform well in solving such kinds of problems. This review discusses areas where metaheuristics are used in the echo state network - a pioneer in the reservoir computing field. In addition, trends and research gaps are also discussed. © 2013 IEEE.