<p><span>Arab users of social media have significantly increased, thus increasing the opportunities for extracting knowledge from various areas of life such as trade, education, psychological health services, etc. The active Arab presence on Twitter motivates many researchers to classify and analysis Arabic tweets from numerous aspects. This study aimed to explore the best performance scenarios in the classification of emotions conveyed through Arabic tweets. Hence, various experiments were conducted to investigate the effects of feature extraction techniques and the N-gram model on the performance of three supervised machine learning algorithms, which are Support Vector Machine (SVM), Naïve Bayes (NB), and Logistic Regression (LR). The general method of the experiments was based on five steps; data collection, preprocessing, feature extraction, emotion classification, and evaluation of results. To implement these experiments, a real-world Twitter dataset was gathered. The best result achieved by the SVM classifier when using a bag of words (BoW) weighting schema (with unigrams and bigrams or with unigrams, bigrams, and trigrams) exceeded the best performance results of other algorithms.</span></p>