Traditional questionnaire-based psychological tests have many drawbacks, such as the subjects' inaccurate understanding of the questions, incomplete answers, disguised intentions, one-way communication, closed answers, etc. This paper proposes a psychological test system design scheme based on pre-trained large language model (LLM), which makes use of LLM's ability to learn and understand natural language and generate output. Complete the psychological test with the subjects through natural language human-computer dialogue.Taking the Cattell 16 Personality Factors Test (16PF) as an example, a natural corpus for large model training of psychological test was researched and compiled, and the psychological test model of two factors in 16PF was developed. Through test comparison, the psychological test model is better than the traditional questionnaire test effect.