Recently, Word Embeddings have been introduced as a major breakthrough in Natural Language Processing (NLP) to learn viable representation of linguistic items based on contextual information or/and word co-occurrence. In this paper, we investigate Arabic document classification using Word and document Embeddings as representational basis rather than relying on text preprocessing and bag-of-words representation. We demonstrate that document Embeddings outperform text preprocessing techniques either by learning them using Doc2Vec or averaging word vectors using a simple method for document Embedding construction. Moreover, the results show that the classification accuracy is less sensitive to word and document vectors learning parameters.