Recommendation systems have become ubiquitous in modern lives. We encounter them when shopping online, streaming movies, or listening to music. Despite the widespread use of these systems, their effect on consumers’ welfare remains ambiguous. Do different types of recommendation systems produce different welfare effects and how do these systems interact with potential behavioral biases? Using the welfare metrics developed in Harrison and Ng (2018) and Alekseev, Harrison, Lau, and Ross (2018), we study the welfare effects of recommendation systems in risky choice. We compare recommendation systems that are based on statistical models and that are based on economic models. We explore whether the decision to follow a recommendation depends on the type of a recommendation system, on whether the logic of a recommendation is transparent, and on the number of alternatives to choose from.