The propensity of consumers to engage in word-of-mouth (WOM) can differ after good versus bad experiences. This can result in positive or negative selection of user-generated reviews. We show how the strength of brand image - determined by the dispersion of consumer beliefs about quality - and the informativeness of good and bad experiences impact the selection of WOM in equilibrium. Our premise is that WOM is costly: Early adopters talk only if their information is instrumental for the receiver’s purchase decision. If the brand image is strong, i.e., consumers have close to homogeneous beliefs about quality, then only negative WOM can arise. With a weak brand image, positive WOM can occur if positive experiences are sufficiently informative. We show that our theoretical predictions are consistent with restaurant review data from Yelp.com. A review rating for a national established chain restaurant is almost 1-star lower (on a 5-star scale) than a review rating for a comparable independent restaurant, controlling for various reviewer and restaurant characteristics. Further, negative chain restaurant reviews have more instances of expectation words, indicating agreement over beliefs about the quality, whereas positive reviews of independent restaurants feature disproportionately many novelty words.