In the present research, the effectiveness of classification algorithms on coral reef habitat mapping was evaluated using Landsat-8 images acquired in 2013. For this purpose, Except Maximum Likelihood algorithm that is common method in coral reef habitat mapping, the efficiency of Neural Network and Support Vector Machine were estimated, as well. Along with data collected from diving in Lizard Island, eastern Australia, the research was accomplished by generalizability of the results to coral reef of Queshm and Larak Islands, Persian Gulf. For this, parallel to getting satellite images, through diving operations, field data were gathered from Queshm and Larak Islands. Depends on the number of classes, the results are varied, so that, Maximum Likelihood has the best efficiency in 2-class classification. However, increasing the number of classes shows more efficiency for Support Vector Machine and Neural Network. In 4-class classification, Support Vector Machine and Neural Network, improve the classification accuracy by 7% and 9% respectively. Implementation of methods in Queshm and Larak Islands shows the generalizability of Support Vector Machine results in this region by 8% improvement in comparison to ML and overall accuracy about 68%, whilst ANN shows the worst results in this region with overall accuracy of 58%, which is because of sensitivity of this algorithm to the number of training data. The capability of SVM to handle mixed pixels and training data deficit issues, cause it to be the best classifier in this case. Finally, because of appropriate performance in both regions and more robustness of results, the SVM by using medium resolution satellite images is selected as the optimized algorithm for mapping of coral reef habitats.