Data on the presence of a number of vegetation 'states' (defined in terms of species dominance in areas of 10×10 cm) and transition probabilities were derived from permanent quadrats in a number of recently burned heath stands. Data were taken from a species-rich community, a species-poor type and a high-level Calluna-Eriophorum bog. Simple Markovian models were constructed using these data, and the model predictions were compared with known or expected trends. Models for species-rich heath yielded poor simulations of expected trends since matrices derived from data for the first years after fire did not contain sufficient information on transitions to states important later in the developmental sequence. Model results for the simpler species-poor and bog communities were more satisfactory and simulated expected trends. In these types all species recovered quickly after fire and less rearrangement of species abundances took place. Maximum likelihood statistics carried out on the transition matrices produced inconclusive results for the species-rich and species-poor types, but indicated that the data from the Calluna-Eriophorum bog approximated a first-order time-homogeneous Markov chain. It was concluded that Markov models lack predictive ability except in relatively simple systems, but that they may be useful in illustrating variations in short-term community dynamics.