I study the optimal design of review systems. A platform learns a product’s quality from user-generated reviews. It faces a trade-off between the informativeness and frequency of reviews. Detailed reviews are individually more informative but less frequently submitted than simple reviews. I characterize the informational content of review systems, which depends on the information in each review and their submission rate. I use this characterization to derive the platform’s optimal review system and show how it varies with reviewers’ preference heterogeneity. In particular, when reviewers are homogeneous the platform isolates “horrible" experiences instead of differentiating between “good" and “bad" experiences.