We consider modelling time series of amounts which may be zero using a stochastic first-order Markov model with mixed transition density having a discrete component at 0 and a continuous component describing non-zero amounts. The models extend chain-dependent stochastic models in the literature on modelling rainfall. Under certain assumptions the Markov chain likelihood can be factored to allow model parameters to be estimated by maximum likelihood using standard Generalized Linear Models methods and software. The results give estimates of seasonal patterns in mean amounts and probability distributions of amounts. We illustrate with 30 years of daily rainfall data from Melbourne, Australia. Copyright © 2000 John Wiley & Sons, Ltd.