Modern online service providers such as online shopping platforms often provide multiple scenarios to meet different user needs. Existing multi-scenario recommendation methods often train a unified model to serve all scenarios. Although much progress has been made, we argue that they ignore the importance of fusing scenario prior knowledge when modeling user’s behaviors as the discrepancy of user’s interests in different scenarios is significant. To solve this problem, we propose a novel Scenario-adaptive Interest Network, which explicitly incorporates scenario-related context into user behavior sequences, allowing for scenario-specific learning of interests refining. SAINet is stacked by a series of Scenario-adaptive Blocks. Each block adaptively integrates scenario-aware context into historical behaviors to distinguish the differences among scenarios and tailor interest representations to match current scenario of each instance. Extensive experiments on real-world datasets and online A/B test demonstrate the superiority of SAINet over state-of-the-art methods.