Based on the daily power load data in Nanjing from 2014 to 2016, using the harmonic analysis and power spectrum method, etc., the study analyzes the change rule and characteristics of power loads in Nanjing, and the relationship between the daily curve of power load and the meteorological elements such as temperature and humidity in extreme power load days. The results show that the seasonal variation of power loads in Nanjing is mainly bimodal. On the intramonth scale, there is a significant 7days period change, which is significantly corresponding to weekly circulation. There are three types of daily extreme load deviation curves: typical bimodal (Ⅰ), unimodal (Ⅱ), and atypical bimodal (Ⅲ). The typesⅠand Ⅲ show significant positive correlation with temperature and hotness index, while all the three types show significant negative correlation with relative humidity. Then, the type Ⅰcorresponds to the circulation controlled by a strong and stable subhigh over Nanjing, weak cold air activity, and small temperature diurnal variation. The typeⅡ happens while Nanjing is affected by frequent cold air activities and the eastwest swing subtropical high, when showers or thunderstorms usually happen in Nanjing. The intensity of the subtropical high corresponding to type Ⅲ is weaker than that to typeⅠ, and the degree of cold air activity is weaker than that to typeⅡ. The stepwise regression method is adopted to establish prediction models for different types of extreme load days. The average relative errors of the prediction models based on the three types are 6.4%, 5.6%, and 5.3%, respectively, which can provide good quantitative power load forecasts for extreme power load days in Nanjing.