Abstract: A new non-parametric method combining a seasonal-trend decomposition procedure loess model, a frequency analysis and a geometrically distributed block Bootstrap method was established to calculate the reference nutrient levels of a shallow lake situated in the east China which is influenced by human activities. Firstly, the best period in which to make observations to assess the reference nutrient levels and their response matter in the lake was calculated out by using a trend-seasonal decomposition model. Then the reference levels of total nitrogen (TN), total phosphorus (TP) and chlorophylla (Chla) were obtained by frequency analysis. Finally, the confidence intervals were calculated using the geometrically distributed block Bootstrap method. This method overcomes the deficiencies of former methods. The reference levels in Taihu Lake, as one typical shallow lake in the east China, were worked out. The results of TN, TP and Chla in Taihu Lake were 0.78 mg/L, 0.030 mg/L and 2.63 μg/L. The 95% confidence intervals were 0.57~0.83 mg/L, 0.025~0.046 mg/L and 1.86~2.65 μg/L respectivly. The reference nutrient levels of other shallow lakes influenced by human activities in east China can also be obtained using this method.