In this paper, we introduce a novel method that combines multiple neural network results to decide the class of the input. This is the first study which used the method for web pages classification. In our model, each element is represented by multiple descriptive images. After the training process of the neural network model, each element is classified by calculating its descriptive image results. We apply our idea to the web page classification problem using Google Image Search results as descriptive images. We obtained a classification rate of 94.90% on the WebScreenshots dataset that contains 20000 web sites in 4 classes. The method is easily applicable to similar problems.