Peatland testate amoebae (TA) are well-established bioindicators for depth to water table (DWT), but effects of hydrological changes on TA communities have never been tested experimentally. We tested this in a field experiment by placing Sphagnum carpets (15 cm diameter) collected in hummock, lawn and pool microsites (origin) at three local conditions (dry, moist and wet) using trenches dug in a peatland. One series of samples was seeded with microorganism extract from all microsites. TA community were analysed at T0: 8-2008, T1: 5-2009 and T2: 8-2009. We analysed the data using conditional inference trees, principal response curves (PRC) and DWT inferred from TA communities using a transfer function used for paleoecological reconstruction. Density declined from T0 to T1 and then increased sharply by T2. Species richness, Simpson diversity and Simpson evenness were lower at T2 than at T0 and T1. Seeded communities had higher species richness in pool samples at T0. Pool samples tended to have higher density, lower species richness, Simpson diversity and Simpson Evenness than hummock and/or lawn samples until T1. In the PRC, the effect of origin was significant at T0 and T1, but the effect faded away by T2. Seeding effect was strongest at T1 and lowest vanished by T2. Local condition effect was strong but not in line with the wetness gradient at T1 but started to reflect it by T2. Likewise, TA-inferred DWT started to match the experimental conditions by T2, but more so in hummock and lawn samples than in pool samples. This study confirmed that TA responds to hydrological changes over a 1-year period. However, sensitivity of TA to hydrological fluctuations, and thus the accuracy of inferred DWT changes, was habitat specific, pool TA communities being least responsive to environmental changes. Lawns and hummocks may be thus better suited than pools for paleoecological reconstructions. This, however, contrasts with the higher prediction error and species' tolerance for DWT with increasing dryness observed in transfer function models.