Sarcasm is an indirect form of speech intentionally used to express a witty remark implicitly rather than its literal meaning. It is a phenomenon very common in social media. Because of its inherently ambiguous nature, detecting sarcasm is a difficult task not only for computers but also for humans. We present a novel computational approach that harnesses the situational and contextual disparity coupled with a sarcasm-indicating lexicon as a basis for sarcasm detection. Our investigations in various factors that indicate situational disparity culminated in Blunt Sentiment Contrast, Word Polarity Contrast, Word Phrase Polarity Contrast and Inter Phrase Polarity Contrast. Alongside, we compiled a lexicon of patterns cuing sarcasm from various sources and assigned weights by 4 different evaluators manually. We also improvised an antecedent bootstrapping algorithm which facilitates automatic learning for contrasting situational phrases. We logically combined the predictions of sarcastic instances given by the feature based SVM classifiers, and those of the pattern matching approach based weighted-lexicon. We then conducted experiments on a collection of 137, 700 tweets collected from Twitter and obtained an average Fscore of 0.93. To validate our approach, we carried out tests on a gold standard test set used in Riloff et al.’s work. We obtained an improvement about 28% over the best results reported in the literature. © 2016, International Journal of Pharmacy and Technology. All rights reserved.