Pristine peatlands are carbon (C)-accumulating wetland ecosystems sustained by a high water table (WT) and consequent anoxia that slows down decomposition. Persistent WT drawdown as a response to climate and/or land-use change affects decomposition either directly through environmental factors such as increased oxygenation, or indirectly through changes in plant community composition. This study attempts to disentangle the direct and indirect effects of WT drawdown by measuring the relative importance of environmental parameters (WT depth, temperature, soil chemistry) and litter type and/or litter chemical quality on the 2-year decomposition rates of above- and belowground litter (altogether 39 litter types). Consequences for organic matter accumulation were estimated based on the annual litter production. The study sites were chosen to form a three-stage chronosequence from pristine (undrained) to short-term (years) and long-term (decades) WT drawdown conditions at three nutrient regimes. The direct effects of WT drawdown were overruled by the indirect effects through changes in litter type composition and production. Short-term responses to WT drawdown were small. In long-term, dramatically increased litter inputs resulted in large accumulation of organic matter in spite of increased decomposition rates. Furthermore, the quality of the accumulated matter greatly changed from that accumulated in pristine conditions. Our results show that the shift in vegetation composition as a response to climate and/or land-use change is the main factor affecting peatland ecosystem C cycle, and thus dynamic vegetation is a necessity in any model applied for estimating responses of C fluxes to changing environment. We provide possible grouping of litter types into plant functional types that the models could utilize. Furthermore, our results clearly show a drop in soil summer temperature as a response to WT drawdown when an initially open peatland converts into a forest ecosystem, which has not yet been considered in the existing models.