Microservice architecture (MSA) is an emerging software architectural style, which differs fundamentally from the monolithic, layered architecture. During the development and maintenance of microservice systems, how to provide an effective service retrieval mechanism is a critical challenge to avoid the problems of rework and duplicate code. Meanwhile, nowadays, using the BDD (Behavior-Driven Development) method to develop microservices becomes more and more popular due to its agility and domain-driven characteristics. BDD is an agile software development approach emphasizing that test cases are written in a common language to include scenarios that describe the features of a target system. In this paper, we propose an approach, referred to as SMSR (Scenario-based MicroService Retrieval), to recommend appropriate microservices to users based on the user-written BDD test scenarios. The proposed service retrieval algorithm is based on word2vec, a widely-used machine learning method in NLP (Natural Language Processing), to perform service filtering and service similarity calculation. Experiment results show that SMSR is able to effectively retrieve appropriate microservices from the service repository.