The distribution of sentiment words plays a crucial role in sentiment analysis. While various machine learning techniques have been proposed for sentiment analysis, limited work has been done to understand the mathematical relationship between sentiment words and the sentiment of a text. Several linguistic laws address the frequency of words in a text, but research literature lacks such laws for sentiment texts. This paper introduces a novel concept, the Entropy-Modulated Sentiment Distribution (EMSD) law, which describes the distribution of sentiment words in a review in terms of entropy and the polarity of the review. The theoretical foundation and mathematical formulation of the law are first discussed. For empirical validation, the law is applied to two different datasets, with parameters calculated based on fivefold average values from the datasets. Finally, a multiple linear regression model is fitted to validate the law on a new sample dataset. The results are statistically validated, and the proposed law is found to effectively model sentiment dynamics in a text.