In e-commerce, retailers or sellers are often assessed by customers or buyers based on reputation information to make wise purchasing decisions. Seller reputation becomes an important credential to shadow seller future behaviour. Most existing reputation models directly aggregate the ratings provided by past buyers. However, it is well documented in practical e-commerce systems that buyers’ ratings can be distorted due to collusion, which negatively affects the applicability of these reputation models. To address this challenging problem, we propose the repurchase-and-return reputation (3R) model, which puts buyers’ ratings into context before aggregating them to compute seller reputation. It considers buyer repurchase and product return behaviour after the point in time when the particular rating was provided. Intuitively, repurchases indicate that the buyers are satisfied with the previously purchased products. Thus, their positive ratings should be given more weight. Similarly, product return behaviours indicate that buyers are dissatisfied with their previous purchasing decisions. Thus, their negative ratings should be given more weight. Based on the proposed 3R reputation model, we design a price premium for a transaction considering the post-purchase behaviour of both the buyer and seller in their transactions. The proposed model is proven capable of achieving a pure strategy Nash equilibrium, in which sellers honestly provide products and buyers prefer to return bad products and repurchase good quality products. Experimental evaluation based on extensive simulation demonstrates that our model can accurately evaluate sellers’ honesty and perform well against prevailing unfair rating attacks.