Classification of high-resolution remote sensing images is a challenging problem. Bag-of-words (BOW) based classification algorithms have obtained good performances in recent years. However, how the procedures of the BOW framework affect the classification result is still an open question. We present three visualization algorithms to reconstruct images from BOW. After visualization, we can see what the computer actually “sees” in an image feature. We also analyze the procedures of the BOW framework, namely, descriptor extraction and histogram generation, in detail. It is found that the descriptors should not be blamed for wrong classification. The histogram generation strategy should be improved to be robust with image transformation. Then some suggestions are posed for future improvement of BOW-based remote sensing image classification algorithms.