Whether it is necessary to reduce nitrogen (N) and/or phosphorus (P) input to mitigate lake eutrophication is controversial. The controversy stems mainly from differences in time and space in previous studies that support the contrasting ideas. To test the response of phytoplankton to various combinations of nutrient control strategies in mesocosms and the possibility of reflecting the conditions in natural ecosystems with short-term experiments, a 9-month experiment was carried out in eight 800-L tanks with four nutrient level combinations (+N+P, -N+P, +N-P, and -N-P), with an 18-month whole-ecosystem experiment in eight similar to 800-m(2) ponds as the reference. Phytoplankton abundance was determined by P not N, regardless of the initial TN/TP level, which was in contrast to the nutrient limitation predicted by the N/P theory. Net natural N inputs were calculated to be 4.9, 6.8, 1.5, and 3.0 g in treatments +N+P, -N+P, +N-P, and -N-P, respectively, suggesting that N deficiency and P addition may promote natural N inputs to support phytoplankton development. However, the compensation process was slow, as suggested by an observed increase in TN after 3 weeks in -N+P and 2 months in -N-P in the tank experiment, and after 3 months in -N+P and similar to 3 months in -N-P in our pond experiment. Obviously, such a slow process cannot be simulated in short-term experiments. The natural N inputs cannot be explained by planktonic N-fixation because N-fixing cyanobacteria were scarce, which was probably because there was a limited pool of species in the tanks. Therefore, based on our results we argue that extrapolating short-term, small-scale experiments to large natural ecosystems does not give reliable, accurate results.