Abstract: Nitrogen concentration is an important indicator for evaluating water quality of rivers and lakes. Its spatio-temporal distribution often varies significantly across different river reaches owing to differences in pollutant source, river flow, self-purification capability, and some other factors. To obtain a comprehensive and in-depth understanding of the water quality in the Songhua River basin, we have developed a latent Dirichlet allocation (LDA) model for analysis on spatio-temporal distribution of nitrogen over this basin based on its observation data of 2006 to 2010. Results showed typical distribution patterns of ammonia nitrogen (NH3-N) and total nitrogen (TN) in the river channels, and their meanings were interpreted and analyzed. Annually, nitrogen concentration took three distribution patterns as typical behaviors at high, medium and low concentration levels respectively. In the whole basin, probabilities of NH3-N distribution in these patterns are 1:3:3 respectively while probabilities of TN distribution are 1:1:1. The distribution of peak nitrogen concentration were also conceptualized into three patterns that frequently occur in dry, normal and wet seasons respectively. On average over the basin, their corresponding probabilities are 6:2:1 for NH3-N and 2:1:1 for TN. In addition, this study revealed that differences in pollution loads of the point source and non-point source are possibly the main cause of the difference in spatio-temporal distributions of nitrogen over different river cross sections.