Abstract Multilayer perceptron approach is a method that can be used to make predictions. The multilayer perceptron includes weighting coefficients which can be determined by different optimization techniques. The weighting coefficients between input layer and hidden layer, also between hidden layer and output layer is the important step for the solution of a multilayer perceptron, and optimized weighting coefficient is used for model predictions. Identifying the weighting coefficients of multilayer perceptron with genetic algorithms is called as geno-multilayer perceptron. In this study, geno-multilayer perceptron approach was used to predict significant wave height. For this purpose, geno-multilayer perceptron approach, a relatively new method, was applied to four stations located in the Lake Okeechobee, Florida, in this study. A comparison between the results of two different training (optimization) algorithms namely genetic algorithms and back propagation algorithms was performed. The prediction results show that optimized (trained) weighting coefficients by genetic algorithms reveal a relatively better agreement with observed data compared to back propagation algorithms. In order to make comparison between observed data and predicted results, statistical indexes including the mean relative error percentages, the mean square errors, the coefficient of efficiency and the chi-square (χ2) parameters were used.