Emotions are an important part of everyday human interaction. Emotions can be expressed by means of written text, verbal speech or facial expressions. In recent years, the practice of expressing emotion in social media or blogs have increased rapidly. People write about their feelings and opinions on any political or global issues. All these social activities have made it essential to gather and analyze human emotion from the text. Although the field of emotion detection has been explored extensively for English language, the investigation of this domain for Bangla language still now in its infant stages. Our paper aims at detecting multi-class emotions from Bangla text using Multinomial Naïve Bayes (NB) classifier along with various features such as stemmer, parts-of-speech (POS) tagger, n-grams, term frequency-inverse document frequency (tf-idf). Our final model was able to classify the text into three emotion classes (happy, sad and angry) with an overall accuracy of 78.6%.