Models were developed for predicting the decomposition of dead wood for the main tree species in Finland, based on data collected from long-term thinning experiments in southern and central Finland. The decomposition rates were strongly related to the number of years after tree death. In contrast to previous studies, which have used the first-order exponential model, we found that the decomposition rate was not constant. Therefore, the Gompertz and Chapman-Richard's functions were fitted to the data. The slow initial decomposition period was mainly due to the fact that most dead trees remained standing as snags after their death. The initial period was followed by a period of rapid decomposition and, finally, by a period of moderately slow decomposition. Birch stems decomposed more rapidly than Scots pine and Norway spruce stems. Decomposition rates of Norway spruce stems were somewhat lower than those of Scots pine. Because the carbon concentration of decaying boles was relatively stable (about 50%) the rate of carbon loss follows that of mass loss. Models were also developed for the probability that a dead tree remains standing as a snag. During the first years after death, the probability was high. Thereafter, it decreased rapidly, the decrease being faster for birch stems than for Scots pine and Norway spruce stems. Almost all stems had fallen down within 40 years after their death. In Scots pine and Norway spruce, most snags remained hard and belonged to decay class 1. In birch, a higher proportion of snags belonged to the more advanced decay classes. The models provide a framework for predicting dead wood dynamics in managed as well as dense unthinned stands. The models can be incorporated into forest management planning systems, thereby facilitating estimates of carbon dynamics.