We develop a sequential learning model to analyse how consumers proactively acquire product quality information through reading online product reviews. Product reviews are often treated as exogenous quality signals in the Internet word-of-mouth literature. Review reading, however, is a costly activity undertaken in a deliberate manner. Our model casts consumers’ review reading process in a rational framework. In our model, whether a consumer reads reviews, and of which products, depend on the consumer’s information set and expectations. We estimate the model using a rich dataset from a restaurant review website which contains information on both browsing and purchase. We find strong evidence of consumers consciously seeking product reviews, and parameter estimates reveal distinct types of information acquisition behaviours. Comparison with alternative models shows that taking reviews as exogenous signals leads to biased estimates on quality levels and signal precisions. Counterfactual analysis further shows that the reviews a consumer sees earlier is more consequential, due partly to their impact on subsequent search actions, and that sorting reviews from lowest to highest ratings leads to highest differentiation across products. Our study is the first to analyse consumer’s review reading process and to analyse the effect of reviews in this dynamic context.