This thesis proposes automatic classification of the emotional content of web documents using Natural Language Processing (NLP) algorithms. We used online articles and general documents to verify the performance of the algorithm, such as general web pages and news articles. The experiments used sentiment analysis that extracts sentiment of web documents. We used unigram and bigram approaches that are known as special types of N-gram, where N=1 and N=2, respectively. The unigram model analyses the probability to hit each word in the corpus independently; however, the bigram model analyses the probability of a word occurring depending on the previous word. Our results show that the unigram model has a better performance compared to the bigram model in terms of automatic classification of the emotional content of web documents.