ABSTRACT Hyper-rectangle (Parallelepiped) supervised classification method is a quick and easy method to implement, with clear definition to each class subspace. Using the training data for each class the limits of the parallelepiped is defined, either by the minimum and maximum pixel values in given class, or by a certain number of standard deviations on either side of the mean of the training data for given class. The pixels lying inside the parallelepipeds are tagged to it. Although the method looks simple and straight forward, but it's very difficult to grant a robust classification performance. The reason is the presence of serious errors that may take place and affect the robustness of the solution. These errors are coming from the possibility of having one or more than one pixel lying in more than one parallelepiped, or fully outside all of them. Basically, theses errors are likely to occur with the more complex feature space. The proposed modification involves applying this method using only one pair of bands at a time to overcome the problem of finding one pixel in more than one class. The new WorldView-2 eight band data will be used in the assessment and verification of the new approach. Moreover, a significant group of unclassified pixels will be tested, which will be classified in a second step using other spectral characteristics such as NDVI and band rationing. The proposed methodology showed a good result in separating between four main classes namely; vegetation, water, shadow, and man-made objects. Conclusions and recommendations are given with respect to the suitability, accuracy and efficiency of this method.