Today, millions of websites on the Internet are widely used to access information. For effective use of web pages with increasing numbers every day, they need to be well classified. In this study, binary and multi-class classification models have been created which can classify web pages with high accuracy. In our experiments, URLs and categories of English web pages in the Open Directory Project (ODP) were used. Training dataset was created by pulling web page texts from URL information. To our knowledge, this is the first comprehensive web page classification dataset for Turkish. In this study, Convolutional Neural Network (CNN), Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) deep learning methods which are effective in text classification are used. Word embedding was used instead of n-gram approaches commonly used for feature extraction in text classification studies. In this study, hyper-parameter optimization was performed for deep learning models. Binary and multi-class classification models were created with the best parameters. Binary classification models were compared with the results of another study, and multi-class classification models were compared with each other. The performances of all models were examined by considering their training time and f1 scores.