The subject of cloud detection has been an interest for many, including geologists, statisticians, and scientists at NASA. With the launch of NASA's Multi-angle Imaging SpectroRadiometer (MISR), we now have access to a vast collection of information to help us identify clouds in the sky. The matter of cloud detection is then transformed into a problem of classification: given a photo and the properties associated with each pixel, how can we tell which pixel is a cloud or not? In this paper, I attempt to classify through a staged approach, first using a generalized linear model to predict, then spatial analysis to further improve where the predicted values fell short. Each stage provides a boost in terms of accuracy in the separation as measured by the error rate.