Quality provision is often low in many developing markets, and firms commonly lack a reputation for quality. This paper examines this issue both theoretically and empirically in the context of retail watermelon markets in China. I first demonstrate the existence of significant asymmetric information on quality between sellers and buyers, as well as the absence of a quality premium at baseline. To explain this, I develop a theoretical model that highlights the role of consumer beliefs and costly signaling in influencing sellers’ reputation incentives. I then conduct an experiment by randomly introducing two signaling technologies into different markets: a cheap sticker label and a more expensive laser-cut label. Consistent with the theoretical predictions, the laser label induces sellers to offer higher quality, resulting in increased sales and profits, while the sticker label fails to achieve the same effect. Using the experimental variation, I estimate an empirical model of consumer learning to uncover the underlying evolution of beliefs. The results show that pessimistic beliefs under the sticker label can hinder reputation building, whereas the laser label enhances consumer learning and strengthens sellers’ reputation incentives.