The study of the residual variation associated with tree diameter growth predictions in one Scots pine ( Pinus sylvestris L.) stand and one Norway spruce ( Picea abies( L.) Karst.) stand is evaluated. Annual diameter growth of trees was smoothed with equations that accounted for the variation caused by predictors commonly used in growth models. A time series of the deviations from the annual smoothing equations was calculated for each tree. The variation in residuals was modelled by four random parameters, which describe the level, trend, autocorrelation and error variance of the time series of the residuals. In growth simulations, the modelled residual variation can be added as a stochastic component to the growth estimate to take into account the total variation in diameter growth, and to obtain a close similarity between real and simulated stand development. In simulations that added the modelled stochastic variation to the diameter growth, the differentiation of trees into diameter classes was more rapid than in deterministic growth simulations. In the stochastic simulations the sawtimber production was greater than in the deterministic ones. The addition of the stochastic residual variation to the growth estimate did not notably affect the volume growth of the stand.