Twitter sentiment analysis became a favorite technique in understanding people’s opinions and attitudes towards various subjects. This study is grounded on the idea that the tremendous amount of Twitter feeds can be a valuable source in gauging people’s perceptions of company brands. As an effort, I propose a novel sentiment analysis approach to evaluate brand equity based on Twitter intended to conceive a more dynamic, costeffective, and automated solution as an alternative to traditional survey methods. Specifically, I (1) generated a brand-relevant opinion lexicon from prior brand research publications and theories, (2) developed a brand-sentiment annotated corpus (N = 5,556) based on the brand-relevant tweets of nine company brands from three different industries, (3) used the corpus to build classification models based on the PMI (a type of lexicon-based) and supervised machine learning methods, and (4) conducted performance evaluation of the models by comparing with the three well-known sentiment analysis methods. The results indicate that my proposed approach provides better performances in brand sentiment classification on Twitter than the prior sentiment analysis methods. There were several interesting findings in which (1) alternatively transformed term-document matrices other than TF such as TF/IDF, PMI-based, and stemmed one did not make any performance improvement in brand evaluation on Twitter; (2) the proposed PMI method proves its usefulness for domain-specific sentiment analysis, displaying its performance comparable to ML methods; (3) the opinion spam handling procedure could have strong influences on the classification performance. This study makes the following contributions. First, I introduce the novel idea of domain-specific sentiment analysis to measure the company brand equity from Twitter and verified its robust performance empirically. Second, the proposed brand-relevant lexicon is grounded on established theories and thus has much potential for further fine-tuning and improvement. Third, the research methodology of sentiment analysis, including opinion spam filtering and brand-relevant lexicon development, could open a path wherein the whole procedure of handling and analyzing a large-scale and messy text data can be fully automated for brand equity evaluation.