 Random forests, RFs, are a popular machine learning algorithm for classification of remote sensing data. This paper examined the effects of various input parameters on the accuracy of RF classifications, including the number of trees in the forest, the size of the training data set, and the use of independent error assessment methods. The authors found that high-dimensional data sets should be reduced prior to classification, and that the size of the training data set should be as large as possible while still representing the true class proportions of the landscape. Additionally, they noted that RFs can be highly sensitive to the size of the training data set, and that it is important to use iterative classifications to assess the stability of the predicted classes. Finally, the authors suggested that the training data should be randomized or created in a way that reflects the actual class proportions of the landscape to ensure accurate predictions. This article was authored by Corrine Millard and Murray Richardson.