Water quality evaluation is an important foundation for the development and utilization of water resources. With the continuous application of artificial intelligence in production and life, water quality evaluation based on artificial neural network has gradually become a research hotspot. Since water quality monitoring data is only data, does not have category attributes, and cannot be directly used as input to neural networks, the water quality data must be firstly classified. In this paper, by constructing a large data set for water quality monitoring, the single-factor index method, principal component analysis method and fuzzy comprehensive evaluation method were used to label the water quality monitoring data by category. The labelling effects of these methods were compared and analysed, and the optimal category labelling scheme was selected to ensure the scientific accuracy of the later neural network water quality evaluation.