I am working on a land cover classification map. I need to divide the ROIs up for a k-class cross validation and am having difficulty determining a way to generate the random sample k-classes themselves without losing the ROI assignment to each individual class. For example, if I have 3 land cover classes that have ROIs created and 500 ROIs total, I want to divide those 500 ROIs into 5 randomly divided groups of 100, use 400 of them for the training of the supervised classifier and the other 100 as the 'ground truth' data for accuracy assessment (repeated so that all the sites are used for training and accuracy). I can export my ROIs as an ASCII, open that in Excel, assign random numbers to the ROIs, divide them into training and 'accuracy' files, and import the new ASCIIs into ENVI, but I cannot figure out how to keep the new random samples divided by class when I reimport them into ENVI. The reimport just creates one ROI class of all of the points. I could create random assignments of each class individually and select a random sample from each class, but that kind of defeats the purpose as the point is to randomly select the points for training and cross validation and not every class has the same number of points. I remember seeing a post on the forums about a similar question previously, but although Google still shows it when I search for help there, I cannot find the entry itself on the ENVI forums since it updated to the new website (even using the search function on the Forums itself), so it appears that it got lost in the migration. Any help/suggestions would be greatly appreciated.