This paper proposes the Large Language Models Enhanced Collaborative Filtering (LLM-CF) framework, which integrates the world knowledge and reasoning capabilities of Large Language Models (LLMs) into Collaborative Filtering (CF) to enhance Recommender Systems (RSs). LLM-CF leverages the in-context learning and chain of thought reasoning of LLMs to distill their knowledge and reasoning into CF. The framework consists of two parts: an offline service part and an online service part. In the offline service, LLMs are fine-tuned to enhance their recommendation capabilities, and CoT reasoning with collaborative filtering information is generated to form an in-context CoT dataset. In the online service, the in-context CoT examples are retrieved, and the world knowledge and reasoning guided CF feature is learned to enhance existing RSs. The proposed LLM-CF framework is efficient and effective, as demonstrated by experiments on three real-world datasets, where it significantly improves the performance of conventional recommendation models in both ranking and retrieval tasks. The framework also shows that LLMs can be effectively used to provide enhanced collaborative filtering information to existing RSs.This paper proposes the Large Language Models Enhanced Collaborative Filtering (LLM-CF) framework, which integrates the world knowledge and reasoning capabilities of Large Language Models (LLMs) into Collaborative Filtering (CF) to enhance Recommender Systems (RSs). LLM-CF leverages the in-context learning and chain of thought reasoning of LLMs to distill their knowledge and reasoning into CF. The framework consists of two parts: an offline service part and an online service part. In the offline service, LLMs are fine-tuned to enhance their recommendation capabilities, and CoT reasoning with collaborative filtering information is generated to form an in-context CoT dataset. In the online service, the in-context CoT examples are retrieved, and the world knowledge and reasoning guided CF feature is learned to enhance existing RSs. The proposed LLM-CF framework is efficient and effective, as demonstrated by experiments on three real-world datasets, where it significantly improves the performance of conventional recommendation models in both ranking and retrieval tasks. The framework also shows that LLMs can be effectively used to provide enhanced collaborative filtering information to existing RSs.